Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlog.c
4 : * PostgreSQL write-ahead log manager
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2020, 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/xlogarchive.h"
37 : #include "access/xloginsert.h"
38 : #include "access/xlogreader.h"
39 : #include "access/xlogutils.h"
40 : #include "catalog/catversion.h"
41 : #include "catalog/pg_control.h"
42 : #include "catalog/pg_database.h"
43 : #include "commands/progress.h"
44 : #include "commands/tablespace.h"
45 : #include "common/controldata_utils.h"
46 : #include "executor/instrument.h"
47 : #include "miscadmin.h"
48 : #include "pg_trace.h"
49 : #include "pgstat.h"
50 : #include "port/atomics.h"
51 : #include "postmaster/bgwriter.h"
52 : #include "postmaster/startup.h"
53 : #include "postmaster/walwriter.h"
54 : #include "replication/basebackup.h"
55 : #include "replication/logical.h"
56 : #include "replication/origin.h"
57 : #include "replication/slot.h"
58 : #include "replication/snapbuild.h"
59 : #include "replication/walreceiver.h"
60 : #include "replication/walsender.h"
61 : #include "storage/bufmgr.h"
62 : #include "storage/fd.h"
63 : #include "storage/ipc.h"
64 : #include "storage/large_object.h"
65 : #include "storage/latch.h"
66 : #include "storage/pmsignal.h"
67 : #include "storage/predicate.h"
68 : #include "storage/proc.h"
69 : #include "storage/procarray.h"
70 : #include "storage/reinit.h"
71 : #include "storage/smgr.h"
72 : #include "storage/spin.h"
73 : #include "storage/sync.h"
74 : #include "utils/builtins.h"
75 : #include "utils/guc.h"
76 : #include "utils/memutils.h"
77 : #include "utils/ps_status.h"
78 : #include "utils/relmapper.h"
79 : #include "utils/snapmgr.h"
80 : #include "utils/timestamp.h"
81 :
82 : extern uint32 bootstrap_data_checksum_version;
83 :
84 : /* Unsupported old recovery command file names (relative to $PGDATA) */
85 : #define RECOVERY_COMMAND_FILE "recovery.conf"
86 : #define RECOVERY_COMMAND_DONE "recovery.done"
87 :
88 : /* User-settable parameters */
89 : int max_wal_size_mb = 1024; /* 1 GB */
90 : int min_wal_size_mb = 80; /* 80 MB */
91 : int wal_keep_segments = 0;
92 : int XLOGbuffers = -1;
93 : int XLogArchiveTimeout = 0;
94 : int XLogArchiveMode = ARCHIVE_MODE_OFF;
95 : char *XLogArchiveCommand = NULL;
96 : bool EnableHotStandby = false;
97 : bool fullPageWrites = true;
98 : bool wal_log_hints = false;
99 : bool wal_compression = false;
100 : char *wal_consistency_checking_string = NULL;
101 : bool *wal_consistency_checking = NULL;
102 : bool wal_init_zero = true;
103 : bool wal_recycle = true;
104 : bool log_checkpoints = false;
105 : int sync_method = DEFAULT_SYNC_METHOD;
106 : int wal_level = WAL_LEVEL_MINIMAL;
107 : int CommitDelay = 0; /* precommit delay in microseconds */
108 : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
109 : int wal_retrieve_retry_interval = 5000;
110 : int max_slot_wal_keep_size_mb = -1;
111 :
112 : #ifdef WAL_DEBUG
113 : bool XLOG_DEBUG = false;
114 : #endif
115 :
116 : int wal_segment_size = DEFAULT_XLOG_SEG_SIZE;
117 :
118 : /*
119 : * Number of WAL insertion locks to use. A higher value allows more insertions
120 : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
121 : * which needs to iterate all the locks.
122 : */
123 : #define NUM_XLOGINSERT_LOCKS 8
124 :
125 : /*
126 : * Max distance from last checkpoint, before triggering a new xlog-based
127 : * checkpoint.
128 : */
129 : int CheckPointSegments;
130 :
131 : /* Estimated distance between checkpoints, in bytes */
132 : static double CheckPointDistanceEstimate = 0;
133 : static double PrevCheckPointDistance = 0;
134 :
135 : /*
136 : * GUC support
137 : */
138 : const struct config_enum_entry sync_method_options[] = {
139 : {"fsync", SYNC_METHOD_FSYNC, false},
140 : #ifdef HAVE_FSYNC_WRITETHROUGH
141 : {"fsync_writethrough", SYNC_METHOD_FSYNC_WRITETHROUGH, false},
142 : #endif
143 : #ifdef HAVE_FDATASYNC
144 : {"fdatasync", SYNC_METHOD_FDATASYNC, false},
145 : #endif
146 : #ifdef OPEN_SYNC_FLAG
147 : {"open_sync", SYNC_METHOD_OPEN, false},
148 : #endif
149 : #ifdef OPEN_DATASYNC_FLAG
150 : {"open_datasync", SYNC_METHOD_OPEN_DSYNC, false},
151 : #endif
152 : {NULL, 0, false}
153 : };
154 :
155 :
156 : /*
157 : * Although only "on", "off", and "always" are documented,
158 : * we accept all the likely variants of "on" and "off".
159 : */
160 : const struct config_enum_entry archive_mode_options[] = {
161 : {"always", ARCHIVE_MODE_ALWAYS, false},
162 : {"on", ARCHIVE_MODE_ON, false},
163 : {"off", ARCHIVE_MODE_OFF, false},
164 : {"true", ARCHIVE_MODE_ON, true},
165 : {"false", ARCHIVE_MODE_OFF, true},
166 : {"yes", ARCHIVE_MODE_ON, true},
167 : {"no", ARCHIVE_MODE_OFF, true},
168 : {"1", ARCHIVE_MODE_ON, true},
169 : {"0", ARCHIVE_MODE_OFF, true},
170 : {NULL, 0, false}
171 : };
172 :
173 : const struct config_enum_entry recovery_target_action_options[] = {
174 : {"pause", RECOVERY_TARGET_ACTION_PAUSE, false},
175 : {"promote", RECOVERY_TARGET_ACTION_PROMOTE, false},
176 : {"shutdown", RECOVERY_TARGET_ACTION_SHUTDOWN, false},
177 : {NULL, 0, false}
178 : };
179 :
180 : /*
181 : * Statistics for current checkpoint are collected in this global struct.
182 : * Because only the checkpointer or a stand-alone backend can perform
183 : * checkpoints, this will be unused in normal backends.
184 : */
185 : CheckpointStatsData CheckpointStats;
186 :
187 : /*
188 : * ThisTimeLineID will be same in all backends --- it identifies current
189 : * WAL timeline for the database system.
190 : */
191 : TimeLineID ThisTimeLineID = 0;
192 :
193 : /*
194 : * Are we doing recovery from XLOG?
195 : *
196 : * This is only ever true in the startup process; it should be read as meaning
197 : * "this process is replaying WAL records", rather than "the system is in
198 : * recovery mode". It should be examined primarily by functions that need
199 : * to act differently when called from a WAL redo function (e.g., to skip WAL
200 : * logging). To check whether the system is in recovery regardless of which
201 : * process you're running in, use RecoveryInProgress() but only after shared
202 : * memory startup and lock initialization.
203 : */
204 : bool InRecovery = false;
205 :
206 : /* Are we in Hot Standby mode? Only valid in startup process, see xlog.h */
207 : HotStandbyState standbyState = STANDBY_DISABLED;
208 :
209 : static XLogRecPtr LastRec;
210 :
211 : /* Local copy of WalRcv->flushedUpto */
212 : static XLogRecPtr flushedUpto = 0;
213 : static TimeLineID receiveTLI = 0;
214 :
215 : /*
216 : * During recovery, lastFullPageWrites keeps track of full_page_writes that
217 : * the replayed WAL records indicate. It's initialized with full_page_writes
218 : * that the recovery starting checkpoint record indicates, and then updated
219 : * each time XLOG_FPW_CHANGE record is replayed.
220 : */
221 : static bool lastFullPageWrites;
222 :
223 : /*
224 : * Local copy of the state tracked by SharedRecoveryState in shared memory,
225 : * It is false if SharedRecoveryState is RECOVERY_STATE_DONE. True actually
226 : * means "not known, need to check the shared state".
227 : */
228 : static bool LocalRecoveryInProgress = true;
229 :
230 : /*
231 : * Local copy of SharedHotStandbyActive variable. False actually means "not
232 : * known, need to check the shared state".
233 : */
234 : static bool LocalHotStandbyActive = false;
235 :
236 : /*
237 : * Local copy of SharedPromoteIsTriggered variable. False actually means "not
238 : * known, need to check the shared state".
239 : */
240 : static bool LocalPromoteIsTriggered = false;
241 :
242 : /*
243 : * Local state for XLogInsertAllowed():
244 : * 1: unconditionally allowed to insert XLOG
245 : * 0: unconditionally not allowed to insert XLOG
246 : * -1: must check RecoveryInProgress(); disallow until it is false
247 : * Most processes start with -1 and transition to 1 after seeing that recovery
248 : * is not in progress. But we can also force the value for special cases.
249 : * The coding in XLogInsertAllowed() depends on the first two of these states
250 : * being numerically the same as bool true and false.
251 : */
252 : static int LocalXLogInsertAllowed = -1;
253 :
254 : /*
255 : * When ArchiveRecoveryRequested is set, archive recovery was requested,
256 : * ie. signal files were present. When InArchiveRecovery is set, we are
257 : * currently recovering using offline XLOG archives. These variables are only
258 : * valid in the startup process.
259 : *
260 : * When ArchiveRecoveryRequested is true, but InArchiveRecovery is false, we're
261 : * currently performing crash recovery using only XLOG files in pg_wal, but
262 : * will switch to using offline XLOG archives as soon as we reach the end of
263 : * WAL in pg_wal.
264 : */
265 : bool ArchiveRecoveryRequested = false;
266 : bool InArchiveRecovery = false;
267 :
268 : static bool standby_signal_file_found = false;
269 : static bool recovery_signal_file_found = false;
270 :
271 : /* Was the last xlog file restored from archive, or local? */
272 : static bool restoredFromArchive = false;
273 :
274 : /* Buffers dedicated to consistency checks of size BLCKSZ */
275 : static char *replay_image_masked = NULL;
276 : static char *master_image_masked = NULL;
277 :
278 : /* options formerly taken from recovery.conf for archive recovery */
279 : char *recoveryRestoreCommand = NULL;
280 : char *recoveryEndCommand = NULL;
281 : char *archiveCleanupCommand = NULL;
282 : RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
283 : bool recoveryTargetInclusive = true;
284 : int recoveryTargetAction = RECOVERY_TARGET_ACTION_PAUSE;
285 : TransactionId recoveryTargetXid;
286 : char *recovery_target_time_string;
287 : static TimestampTz recoveryTargetTime;
288 : const char *recoveryTargetName;
289 : XLogRecPtr recoveryTargetLSN;
290 : int recovery_min_apply_delay = 0;
291 :
292 : /* options formerly taken from recovery.conf for XLOG streaming */
293 : bool StandbyModeRequested = false;
294 : char *PrimaryConnInfo = NULL;
295 : char *PrimarySlotName = NULL;
296 : char *PromoteTriggerFile = NULL;
297 : bool wal_receiver_create_temp_slot = false;
298 :
299 : /* are we currently in standby mode? */
300 : bool StandbyMode = false;
301 :
302 : /* whether request for fast promotion has been made yet */
303 : static bool fast_promote = false;
304 :
305 : /*
306 : * if recoveryStopsBefore/After returns true, it saves information of the stop
307 : * point here
308 : */
309 : static TransactionId recoveryStopXid;
310 : static TimestampTz recoveryStopTime;
311 : static XLogRecPtr recoveryStopLSN;
312 : static char recoveryStopName[MAXFNAMELEN];
313 : static bool recoveryStopAfter;
314 :
315 : /*
316 : * During normal operation, the only timeline we care about is ThisTimeLineID.
317 : * During recovery, however, things are more complicated. To simplify life
318 : * for rmgr code, we keep ThisTimeLineID set to the "current" timeline as we
319 : * scan through the WAL history (that is, it is the line that was active when
320 : * the currently-scanned WAL record was generated). We also need these
321 : * timeline values:
322 : *
323 : * recoveryTargetTimeLineGoal: what the user requested, if any
324 : *
325 : * recoveryTargetTLIRequested: numeric value of requested timeline, if constant
326 : *
327 : * recoveryTargetTLI: the currently understood target timeline; changes
328 : *
329 : * expectedTLEs: a list of TimeLineHistoryEntries for recoveryTargetTLI and the timelines of
330 : * its known parents, newest first (so recoveryTargetTLI is always the
331 : * first list member). Only these TLIs are expected to be seen in the WAL
332 : * segments we read, and indeed only these TLIs will be considered as
333 : * candidate WAL files to open at all.
334 : *
335 : * curFileTLI: the TLI appearing in the name of the current input WAL file.
336 : * (This is not necessarily the same as ThisTimeLineID, because we could
337 : * be scanning data that was copied from an ancestor timeline when the current
338 : * file was created.) During a sequential scan we do not allow this value
339 : * to decrease.
340 : */
341 : RecoveryTargetTimeLineGoal recoveryTargetTimeLineGoal = RECOVERY_TARGET_TIMELINE_LATEST;
342 : TimeLineID recoveryTargetTLIRequested = 0;
343 : TimeLineID recoveryTargetTLI = 0;
344 : static List *expectedTLEs;
345 : static TimeLineID curFileTLI;
346 :
347 : /*
348 : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
349 : * current backend. It is updated for all inserts. XactLastRecEnd points to
350 : * end+1 of the last record, and is reset when we end a top-level transaction,
351 : * or start a new one; so it can be used to tell if the current transaction has
352 : * created any XLOG records.
353 : *
354 : * While in parallel mode, this may not be fully up to date. When committing,
355 : * a transaction can assume this covers all xlog records written either by the
356 : * user backend or by any parallel worker which was present at any point during
357 : * the transaction. But when aborting, or when still in parallel mode, other
358 : * parallel backends may have written WAL records at later LSNs than the value
359 : * stored here. The parallel leader advances its own copy, when necessary,
360 : * in WaitForParallelWorkersToFinish.
361 : */
362 : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
363 : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
364 : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
365 :
366 : /*
367 : * RedoRecPtr is this backend's local copy of the REDO record pointer
368 : * (which is almost but not quite the same as a pointer to the most recent
369 : * CHECKPOINT record). We update this from the shared-memory copy,
370 : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
371 : * hold an insertion lock). See XLogInsertRecord for details. We are also
372 : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
373 : * see GetRedoRecPtr. A freshly spawned backend obtains the value during
374 : * InitXLOGAccess.
375 : */
376 : static XLogRecPtr RedoRecPtr;
377 :
378 : /*
379 : * doPageWrites is this backend's local copy of (forcePageWrites ||
380 : * fullPageWrites). It is used together with RedoRecPtr to decide whether
381 : * a full-page image of a page need to be taken.
382 : */
383 : static bool doPageWrites;
384 :
385 : /* Has the recovery code requested a walreceiver wakeup? */
386 : static bool doRequestWalReceiverReply;
387 :
388 : /*
389 : * RedoStartLSN points to the checkpoint's REDO location which is specified
390 : * in a backup label file, backup history file or control file. In standby
391 : * mode, XLOG streaming usually starts from the position where an invalid
392 : * record was found. But if we fail to read even the initial checkpoint
393 : * record, we use the REDO location instead of the checkpoint location as
394 : * the start position of XLOG streaming. Otherwise we would have to jump
395 : * backwards to the REDO location after reading the checkpoint record,
396 : * because the REDO record can precede the checkpoint record.
397 : */
398 : static XLogRecPtr RedoStartLSN = InvalidXLogRecPtr;
399 :
400 : /*----------
401 : * Shared-memory data structures for XLOG control
402 : *
403 : * LogwrtRqst indicates a byte position that we need to write and/or fsync
404 : * the log up to (all records before that point must be written or fsynced).
405 : * LogwrtResult indicates the byte positions we have already written/fsynced.
406 : * These structs are identical but are declared separately to indicate their
407 : * slightly different functions.
408 : *
409 : * To read XLogCtl->LogwrtResult, you must hold either info_lck or
410 : * WALWriteLock. To update it, you need to hold both locks. The point of
411 : * this arrangement is that the value can be examined by code that already
412 : * holds WALWriteLock without needing to grab info_lck as well. In addition
413 : * to the shared variable, each backend has a private copy of LogwrtResult,
414 : * which is updated when convenient.
415 : *
416 : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
417 : * (protected by info_lck), but we don't need to cache any copies of it.
418 : *
419 : * info_lck is only held long enough to read/update the protected variables,
420 : * so it's a plain spinlock. The other locks are held longer (potentially
421 : * over I/O operations), so we use LWLocks for them. These locks are:
422 : *
423 : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
424 : * It is only held while initializing and changing the mapping. If the
425 : * contents of the buffer being replaced haven't been written yet, the mapping
426 : * lock is released while the write is done, and reacquired afterwards.
427 : *
428 : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
429 : * XLogFlush).
430 : *
431 : * ControlFileLock: must be held to read/update control file or create
432 : * new log file.
433 : *
434 : * CheckpointLock: must be held to do a checkpoint or restartpoint (ensures
435 : * only one checkpointer at a time; currently, with all checkpoints done by
436 : * the checkpointer, this is just pro forma).
437 : *
438 : *----------
439 : */
440 :
441 : typedef struct XLogwrtRqst
442 : {
443 : XLogRecPtr Write; /* last byte + 1 to write out */
444 : XLogRecPtr Flush; /* last byte + 1 to flush */
445 : } XLogwrtRqst;
446 :
447 : typedef struct XLogwrtResult
448 : {
449 : XLogRecPtr Write; /* last byte + 1 written out */
450 : XLogRecPtr Flush; /* last byte + 1 flushed */
451 : } XLogwrtResult;
452 :
453 : /*
454 : * Inserting to WAL is protected by a small fixed number of WAL insertion
455 : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
456 : * matter which one. To lock out other concurrent insertions, you must hold
457 : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
458 : * indicator of how far the insertion has progressed (insertingAt).
459 : *
460 : * The insertingAt values are read when a process wants to flush WAL from
461 : * the in-memory buffers to disk, to check that all the insertions to the
462 : * region the process is about to write out have finished. You could simply
463 : * wait for all currently in-progress insertions to finish, but the
464 : * insertingAt indicator allows you to ignore insertions to later in the WAL,
465 : * so that you only wait for the insertions that are modifying the buffers
466 : * you're about to write out.
467 : *
468 : * This isn't just an optimization. If all the WAL buffers are dirty, an
469 : * inserter that's holding a WAL insert lock might need to evict an old WAL
470 : * buffer, which requires flushing the WAL. If it's possible for an inserter
471 : * to block on another inserter unnecessarily, deadlock can arise when two
472 : * inserters holding a WAL insert lock wait for each other to finish their
473 : * insertion.
474 : *
475 : * Small WAL records that don't cross a page boundary never update the value,
476 : * the WAL record is just copied to the page and the lock is released. But
477 : * to avoid the deadlock-scenario explained above, the indicator is always
478 : * updated before sleeping while holding an insertion lock.
479 : *
480 : * lastImportantAt contains the LSN of the last important WAL record inserted
481 : * using a given lock. This value is used to detect if there has been
482 : * important WAL activity since the last time some action, like a checkpoint,
483 : * was performed - allowing to not repeat the action if not. The LSN is
484 : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
485 : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
486 : * records. Tracking the WAL activity directly in WALInsertLock has the
487 : * advantage of not needing any additional locks to update the value.
488 : */
489 : typedef struct
490 : {
491 : LWLock lock;
492 : XLogRecPtr insertingAt;
493 : XLogRecPtr lastImportantAt;
494 : } WALInsertLock;
495 :
496 : /*
497 : * All the WAL insertion locks are allocated as an array in shared memory. We
498 : * force the array stride to be a power of 2, which saves a few cycles in
499 : * indexing, but more importantly also ensures that individual slots don't
500 : * cross cache line boundaries. (Of course, we have to also ensure that the
501 : * array start address is suitably aligned.)
502 : */
503 : typedef union WALInsertLockPadded
504 : {
505 : WALInsertLock l;
506 : char pad[PG_CACHE_LINE_SIZE];
507 : } WALInsertLockPadded;
508 :
509 : /*
510 : * State of an exclusive backup, necessary to control concurrent activities
511 : * across sessions when working on exclusive backups.
512 : *
513 : * EXCLUSIVE_BACKUP_NONE means that there is no exclusive backup actually
514 : * running, to be more precise pg_start_backup() is not being executed for
515 : * an exclusive backup and there is no exclusive backup in progress.
516 : * EXCLUSIVE_BACKUP_STARTING means that pg_start_backup() is starting an
517 : * exclusive backup.
518 : * EXCLUSIVE_BACKUP_IN_PROGRESS means that pg_start_backup() has finished
519 : * running and an exclusive backup is in progress. pg_stop_backup() is
520 : * needed to finish it.
521 : * EXCLUSIVE_BACKUP_STOPPING means that pg_stop_backup() is stopping an
522 : * exclusive backup.
523 : */
524 : typedef enum ExclusiveBackupState
525 : {
526 : EXCLUSIVE_BACKUP_NONE = 0,
527 : EXCLUSIVE_BACKUP_STARTING,
528 : EXCLUSIVE_BACKUP_IN_PROGRESS,
529 : EXCLUSIVE_BACKUP_STOPPING
530 : } ExclusiveBackupState;
531 :
532 : /*
533 : * Session status of running backup, used for sanity checks in SQL-callable
534 : * functions to start and stop backups.
535 : */
536 : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
537 :
538 : /*
539 : * Shared state data for WAL insertion.
540 : */
541 : typedef struct XLogCtlInsert
542 : {
543 : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
544 :
545 : /*
546 : * CurrBytePos is the end of reserved WAL. The next record will be
547 : * inserted at that position. PrevBytePos is the start position of the
548 : * previously inserted (or rather, reserved) record - it is copied to the
549 : * prev-link of the next record. These are stored as "usable byte
550 : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
551 : */
552 : uint64 CurrBytePos;
553 : uint64 PrevBytePos;
554 :
555 : /*
556 : * Make sure the above heavily-contended spinlock and byte positions are
557 : * on their own cache line. In particular, the RedoRecPtr and full page
558 : * write variables below should be on a different cache line. They are
559 : * read on every WAL insertion, but updated rarely, and we don't want
560 : * those reads to steal the cache line containing Curr/PrevBytePos.
561 : */
562 : char pad[PG_CACHE_LINE_SIZE];
563 :
564 : /*
565 : * fullPageWrites is the master copy used by all backends to determine
566 : * whether to write full-page to WAL, instead of using process-local one.
567 : * This is required because, when full_page_writes is changed by SIGHUP,
568 : * we must WAL-log it before it actually affects WAL-logging by backends.
569 : * Checkpointer sets at startup or after SIGHUP.
570 : *
571 : * To read these fields, you must hold an insertion lock. To modify them,
572 : * you must hold ALL the locks.
573 : */
574 : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
575 : bool forcePageWrites; /* forcing full-page writes for PITR? */
576 : bool fullPageWrites;
577 :
578 : /*
579 : * exclusiveBackupState indicates the state of an exclusive backup (see
580 : * comments of ExclusiveBackupState for more details). nonExclusiveBackups
581 : * is a counter indicating the number of streaming base backups currently
582 : * in progress. forcePageWrites is set to true when either of these is
583 : * non-zero. lastBackupStart is the latest checkpoint redo location used
584 : * as a starting point for an online backup.
585 : */
586 : ExclusiveBackupState exclusiveBackupState;
587 : int nonExclusiveBackups;
588 : XLogRecPtr lastBackupStart;
589 :
590 : /*
591 : * WAL insertion locks.
592 : */
593 : WALInsertLockPadded *WALInsertLocks;
594 : } XLogCtlInsert;
595 :
596 : /*
597 : * Total shared-memory state for XLOG.
598 : */
599 : typedef struct XLogCtlData
600 : {
601 : XLogCtlInsert Insert;
602 :
603 : /* Protected by info_lck: */
604 : XLogwrtRqst LogwrtRqst;
605 : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
606 : FullTransactionId ckptFullXid; /* nextFullXid of latest checkpoint */
607 : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
608 : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
609 :
610 : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
611 :
612 : /* Fake LSN counter, for unlogged relations. Protected by ulsn_lck. */
613 : XLogRecPtr unloggedLSN;
614 : slock_t ulsn_lck;
615 :
616 : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
617 : pg_time_t lastSegSwitchTime;
618 : XLogRecPtr lastSegSwitchLSN;
619 :
620 : /*
621 : * Protected by info_lck and WALWriteLock (you must hold either lock to
622 : * read it, but both to update)
623 : */
624 : XLogwrtResult LogwrtResult;
625 :
626 : /*
627 : * Latest initialized page in the cache (last byte position + 1).
628 : *
629 : * To change the identity of a buffer (and InitializedUpTo), you need to
630 : * hold WALBufMappingLock. To change the identity of a buffer that's
631 : * still dirty, the old page needs to be written out first, and for that
632 : * you need WALWriteLock, and you need to ensure that there are no
633 : * in-progress insertions to the page by calling
634 : * WaitXLogInsertionsToFinish().
635 : */
636 : XLogRecPtr InitializedUpTo;
637 :
638 : /*
639 : * These values do not change after startup, although the pointed-to pages
640 : * and xlblocks values certainly do. xlblocks values are protected by
641 : * WALBufMappingLock.
642 : */
643 : char *pages; /* buffers for unwritten XLOG pages */
644 : XLogRecPtr *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
645 : int XLogCacheBlck; /* highest allocated xlog buffer index */
646 :
647 : /*
648 : * Shared copy of ThisTimeLineID. Does not change after end-of-recovery.
649 : * If we created a new timeline when the system was started up,
650 : * PrevTimeLineID is the old timeline's ID that we forked off from.
651 : * Otherwise it's equal to ThisTimeLineID.
652 : */
653 : TimeLineID ThisTimeLineID;
654 : TimeLineID PrevTimeLineID;
655 :
656 : /*
657 : * SharedRecoveryState indicates if we're still in crash or archive
658 : * recovery. Protected by info_lck.
659 : */
660 : RecoveryState SharedRecoveryState;
661 :
662 : /*
663 : * SharedHotStandbyActive indicates if we allow hot standby queries to be
664 : * run. Protected by info_lck.
665 : */
666 : bool SharedHotStandbyActive;
667 :
668 : /*
669 : * SharedPromoteIsTriggered indicates if a standby promotion has been
670 : * triggered. Protected by info_lck.
671 : */
672 : bool SharedPromoteIsTriggered;
673 :
674 : /*
675 : * WalWriterSleeping indicates whether the WAL writer is currently in
676 : * low-power mode (and hence should be nudged if an async commit occurs).
677 : * Protected by info_lck.
678 : */
679 : bool WalWriterSleeping;
680 :
681 : /*
682 : * recoveryWakeupLatch is used to wake up the startup process to continue
683 : * WAL replay, if it is waiting for WAL to arrive or failover trigger file
684 : * to appear.
685 : */
686 : Latch recoveryWakeupLatch;
687 :
688 : /*
689 : * During recovery, we keep a copy of the latest checkpoint record here.
690 : * lastCheckPointRecPtr points to start of checkpoint record and
691 : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
692 : * checkpointer when it wants to create a restartpoint.
693 : *
694 : * Protected by info_lck.
695 : */
696 : XLogRecPtr lastCheckPointRecPtr;
697 : XLogRecPtr lastCheckPointEndPtr;
698 : CheckPoint lastCheckPoint;
699 :
700 : /*
701 : * lastReplayedEndRecPtr points to end+1 of the last record successfully
702 : * replayed. When we're currently replaying a record, ie. in a redo
703 : * function, replayEndRecPtr points to the end+1 of the record being
704 : * replayed, otherwise it's equal to lastReplayedEndRecPtr.
705 : */
706 : XLogRecPtr lastReplayedEndRecPtr;
707 : TimeLineID lastReplayedTLI;
708 : XLogRecPtr replayEndRecPtr;
709 : TimeLineID replayEndTLI;
710 : /* timestamp of last COMMIT/ABORT record replayed (or being replayed) */
711 : TimestampTz recoveryLastXTime;
712 :
713 : /*
714 : * timestamp of when we started replaying the current chunk of WAL data,
715 : * only relevant for replication or archive recovery
716 : */
717 : TimestampTz currentChunkStartTime;
718 : /* Are we requested to pause recovery? */
719 : bool recoveryPause;
720 :
721 : /*
722 : * lastFpwDisableRecPtr points to the start of the last replayed
723 : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
724 : */
725 : XLogRecPtr lastFpwDisableRecPtr;
726 :
727 : slock_t info_lck; /* locks shared variables shown above */
728 : } XLogCtlData;
729 :
730 : static XLogCtlData *XLogCtl = NULL;
731 :
732 : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
733 : static WALInsertLockPadded *WALInsertLocks = NULL;
734 :
735 : /*
736 : * We maintain an image of pg_control in shared memory.
737 : */
738 : static ControlFileData *ControlFile = NULL;
739 :
740 : /*
741 : * Calculate the amount of space left on the page after 'endptr'. Beware
742 : * multiple evaluation!
743 : */
744 : #define INSERT_FREESPACE(endptr) \
745 : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
746 :
747 : /* Macro to advance to next buffer index. */
748 : #define NextBufIdx(idx) \
749 : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
750 :
751 : /*
752 : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
753 : * would hold if it was in cache, the page containing 'recptr'.
754 : */
755 : #define XLogRecPtrToBufIdx(recptr) \
756 : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
757 :
758 : /*
759 : * These are the number of bytes in a WAL page usable for WAL data.
760 : */
761 : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
762 :
763 : /* Convert values of GUCs measured in megabytes to equiv. segment count */
764 : #define ConvertToXSegs(x, segsize) \
765 : (x / ((segsize) / (1024 * 1024)))
766 :
767 : /* The number of bytes in a WAL segment usable for WAL data. */
768 : static int UsableBytesInSegment;
769 :
770 : /*
771 : * Private, possibly out-of-date copy of shared LogwrtResult.
772 : * See discussion above.
773 : */
774 : static XLogwrtResult LogwrtResult = {0, 0};
775 :
776 : /*
777 : * Codes indicating where we got a WAL file from during recovery, or where
778 : * to attempt to get one.
779 : */
780 : typedef enum
781 : {
782 : XLOG_FROM_ANY = 0, /* request to read WAL from any source */
783 : XLOG_FROM_ARCHIVE, /* restored using restore_command */
784 : XLOG_FROM_PG_WAL, /* existing file in pg_wal */
785 : XLOG_FROM_STREAM /* streamed from master */
786 : } XLogSource;
787 :
788 : /* human-readable names for XLogSources, for debugging output */
789 : static const char *const xlogSourceNames[] = {"any", "archive", "pg_wal", "stream"};
790 :
791 : /*
792 : * openLogFile is -1 or a kernel FD for an open log file segment.
793 : * openLogSegNo identifies the segment. These variables are only used to
794 : * write the XLOG, and so will normally refer to the active segment.
795 : * Note: call Reserve/ReleaseExternalFD to track consumption of this FD.
796 : */
797 : static int openLogFile = -1;
798 : static XLogSegNo openLogSegNo = 0;
799 :
800 : /*
801 : * These variables are used similarly to the ones above, but for reading
802 : * the XLOG. Note, however, that readOff generally represents the offset
803 : * of the page just read, not the seek position of the FD itself, which
804 : * will be just past that page. readLen indicates how much of the current
805 : * page has been read into readBuf, and readSource indicates where we got
806 : * the currently open file from.
807 : * Note: we could use Reserve/ReleaseExternalFD to track consumption of
808 : * this FD too; but it doesn't currently seem worthwhile, since the XLOG is
809 : * not read by general-purpose sessions.
810 : */
811 : static int readFile = -1;
812 : static XLogSegNo readSegNo = 0;
813 : static uint32 readOff = 0;
814 : static uint32 readLen = 0;
815 : static XLogSource readSource = XLOG_FROM_ANY;
816 :
817 : /*
818 : * Keeps track of which source we're currently reading from. This is
819 : * different from readSource in that this is always set, even when we don't
820 : * currently have a WAL file open. If lastSourceFailed is set, our last
821 : * attempt to read from currentSource failed, and we should try another source
822 : * next.
823 : *
824 : * pendingWalRcvRestart is set when a config change occurs that requires a
825 : * walreceiver restart. This is only valid in XLOG_FROM_STREAM state.
826 : */
827 : static XLogSource currentSource = XLOG_FROM_ANY;
828 : static bool lastSourceFailed = false;
829 : static bool pendingWalRcvRestart = false;
830 :
831 : typedef struct XLogPageReadPrivate
832 : {
833 : int emode;
834 : bool fetching_ckpt; /* are we fetching a checkpoint record? */
835 : bool randAccess;
836 : } XLogPageReadPrivate;
837 :
838 : /*
839 : * These variables track when we last obtained some WAL data to process,
840 : * and where we got it from. (XLogReceiptSource is initially the same as
841 : * readSource, but readSource gets reset to zero when we don't have data
842 : * to process right now. It is also different from currentSource, which
843 : * also changes when we try to read from a source and fail, while
844 : * XLogReceiptSource tracks where we last successfully read some WAL.)
845 : */
846 : static TimestampTz XLogReceiptTime = 0;
847 : static XLogSource XLogReceiptSource = XLOG_FROM_ANY;
848 :
849 : /* State information for XLOG reading */
850 : static XLogRecPtr ReadRecPtr; /* start of last record read */
851 : static XLogRecPtr EndRecPtr; /* end+1 of last record read */
852 :
853 : /*
854 : * Local copies of equivalent fields in the control file. When running
855 : * crash recovery, minRecoveryPoint is set to InvalidXLogRecPtr as we
856 : * expect to replay all the WAL available, and updateMinRecoveryPoint is
857 : * switched to false to prevent any updates while replaying records.
858 : * Those values are kept consistent as long as crash recovery runs.
859 : */
860 : static XLogRecPtr minRecoveryPoint;
861 : static TimeLineID minRecoveryPointTLI;
862 : static bool updateMinRecoveryPoint = true;
863 :
864 : /*
865 : * Have we reached a consistent database state? In crash recovery, we have
866 : * to replay all the WAL, so reachedConsistency is never set. During archive
867 : * recovery, the database is consistent once minRecoveryPoint is reached.
868 : */
869 : bool reachedConsistency = false;
870 :
871 : static bool InRedo = false;
872 :
873 : /* Have we launched bgwriter during recovery? */
874 : static bool bgwriterLaunched = false;
875 :
876 : /* For WALInsertLockAcquire/Release functions */
877 : static int MyLockNo = 0;
878 : static bool holdingAllLocks = false;
879 :
880 : #ifdef WAL_DEBUG
881 : static MemoryContext walDebugCxt = NULL;
882 : #endif
883 :
884 : static void readRecoverySignalFile(void);
885 : static void validateRecoveryParameters(void);
886 : static void exitArchiveRecovery(TimeLineID endTLI, XLogRecPtr endOfLog);
887 : static bool recoveryStopsBefore(XLogReaderState *record);
888 : static bool recoveryStopsAfter(XLogReaderState *record);
889 : static void recoveryPausesHere(bool endOfRecovery);
890 : static bool recoveryApplyDelay(XLogReaderState *record);
891 : static void SetLatestXTime(TimestampTz xtime);
892 : static void SetCurrentChunkStartTime(TimestampTz xtime);
893 : static void CheckRequiredParameterValues(void);
894 : static void XLogReportParameters(void);
895 : static void checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI,
896 : TimeLineID prevTLI);
897 : static void LocalSetXLogInsertAllowed(void);
898 : static void CreateEndOfRecoveryRecord(void);
899 : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
900 : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
901 : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
902 :
903 : static void AdvanceXLInsertBuffer(XLogRecPtr upto, bool opportunistic);
904 : static bool XLogCheckpointNeeded(XLogSegNo new_segno);
905 : static void XLogWrite(XLogwrtRqst WriteRqst, bool flexible);
906 : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
907 : bool find_free, XLogSegNo max_segno,
908 : bool use_lock);
909 : static int XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
910 : XLogSource source, bool notfoundOk);
911 : static int XLogFileReadAnyTLI(XLogSegNo segno, int emode, XLogSource source);
912 : static int XLogPageRead(XLogReaderState *xlogreader, XLogRecPtr targetPagePtr,
913 : int reqLen, XLogRecPtr targetRecPtr, char *readBuf);
914 : static bool WaitForWALToBecomeAvailable(XLogRecPtr RecPtr, bool randAccess,
915 : bool fetching_ckpt, XLogRecPtr tliRecPtr);
916 : static int emode_for_corrupt_record(int emode, XLogRecPtr RecPtr);
917 : static void XLogFileClose(void);
918 : static void PreallocXlogFiles(XLogRecPtr endptr);
919 : static void RemoveTempXlogFiles(void);
920 : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr);
921 : static void RemoveXlogFile(const char *segname, XLogRecPtr lastredoptr, XLogRecPtr endptr);
922 : static void UpdateLastRemovedPtr(char *filename);
923 : static void ValidateXLOGDirectoryStructure(void);
924 : static void CleanupBackupHistory(void);
925 : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
926 : static XLogRecord *ReadRecord(XLogReaderState *xlogreader,
927 : int emode, bool fetching_ckpt);
928 : static void CheckRecoveryConsistency(void);
929 : static XLogRecord *ReadCheckpointRecord(XLogReaderState *xlogreader,
930 : XLogRecPtr RecPtr, int whichChkpt, bool report);
931 : static bool rescanLatestTimeLine(void);
932 : static void InitControlFile(uint64 sysidentifier);
933 : static void WriteControlFile(void);
934 : static void ReadControlFile(void);
935 : static char *str_time(pg_time_t tnow);
936 : static void SetPromoteIsTriggered(void);
937 : static bool CheckForStandbyTrigger(void);
938 :
939 : #ifdef WAL_DEBUG
940 : static void xlog_outrec(StringInfo buf, XLogReaderState *record);
941 : #endif
942 : static void xlog_outdesc(StringInfo buf, XLogReaderState *record);
943 : static void pg_start_backup_callback(int code, Datum arg);
944 : static void pg_stop_backup_callback(int code, Datum arg);
945 : static bool read_backup_label(XLogRecPtr *checkPointLoc,
946 : bool *backupEndRequired, bool *backupFromStandby);
947 : static bool read_tablespace_map(List **tablespaces);
948 :
949 : static void rm_redo_error_callback(void *arg);
950 : static int get_sync_bit(int method);
951 :
952 : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
953 : XLogRecData *rdata,
954 : XLogRecPtr StartPos, XLogRecPtr EndPos);
955 : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
956 : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
957 : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
958 : XLogRecPtr *PrevPtr);
959 : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
960 : static char *GetXLogBuffer(XLogRecPtr ptr);
961 : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
962 : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
963 : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
964 : static void checkXLogConsistency(XLogReaderState *record);
965 :
966 : static void WALInsertLockAcquire(void);
967 : static void WALInsertLockAcquireExclusive(void);
968 : static void WALInsertLockRelease(void);
969 : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
970 :
971 : /*
972 : * Insert an XLOG record represented by an already-constructed chain of data
973 : * chunks. This is a low-level routine; to construct the WAL record header
974 : * and data, use the higher-level routines in xloginsert.c.
975 : *
976 : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
977 : * WAL record applies to, that were not included in the record as full page
978 : * images. If fpw_lsn <= RedoRecPtr, the function does not perform the
979 : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
980 : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
981 : * record is always inserted.
982 : *
983 : * 'flags' gives more in-depth control on the record being inserted. See
984 : * XLogSetRecordFlags() for details.
985 : *
986 : * The first XLogRecData in the chain must be for the record header, and its
987 : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
988 : * xl_crc fields in the header, the rest of the header must already be filled
989 : * by the caller.
990 : *
991 : * Returns XLOG pointer to end of record (beginning of next record).
992 : * This can be used as LSN for data pages affected by the logged action.
993 : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
994 : * before the data page can be written out. This implements the basic
995 : * WAL rule "write the log before the data".)
996 : */
997 : XLogRecPtr
998 29389778 : XLogInsertRecord(XLogRecData *rdata,
999 : XLogRecPtr fpw_lsn,
1000 : uint8 flags,
1001 : int num_fpi)
1002 : {
1003 29389778 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1004 : pg_crc32c rdata_crc;
1005 : bool inserted;
1006 29389778 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
1007 29389778 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
1008 29389778 : bool isLogSwitch = (rechdr->xl_rmid == RM_XLOG_ID &&
1009 : info == XLOG_SWITCH);
1010 : XLogRecPtr StartPos;
1011 : XLogRecPtr EndPos;
1012 29389778 : bool prevDoPageWrites = doPageWrites;
1013 :
1014 : /* we assume that all of the record header is in the first chunk */
1015 : Assert(rdata->len >= SizeOfXLogRecord);
1016 :
1017 : /* cross-check on whether we should be here or not */
1018 29389778 : if (!XLogInsertAllowed())
1019 0 : elog(ERROR, "cannot make new WAL entries during recovery");
1020 :
1021 : /*----------
1022 : *
1023 : * We have now done all the preparatory work we can without holding a
1024 : * lock or modifying shared state. From here on, inserting the new WAL
1025 : * record to the shared WAL buffer cache is a two-step process:
1026 : *
1027 : * 1. Reserve the right amount of space from the WAL. The current head of
1028 : * reserved space is kept in Insert->CurrBytePos, and is protected by
1029 : * insertpos_lck.
1030 : *
1031 : * 2. Copy the record to the reserved WAL space. This involves finding the
1032 : * correct WAL buffer containing the reserved space, and copying the
1033 : * record in place. This can be done concurrently in multiple processes.
1034 : *
1035 : * To keep track of which insertions are still in-progress, each concurrent
1036 : * inserter acquires an insertion lock. In addition to just indicating that
1037 : * an insertion is in progress, the lock tells others how far the inserter
1038 : * has progressed. There is a small fixed number of insertion locks,
1039 : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
1040 : * boundary, it updates the value stored in the lock to the how far it has
1041 : * inserted, to allow the previous buffer to be flushed.
1042 : *
1043 : * Holding onto an insertion lock also protects RedoRecPtr and
1044 : * fullPageWrites from changing until the insertion is finished.
1045 : *
1046 : * Step 2 can usually be done completely in parallel. If the required WAL
1047 : * page is not initialized yet, you have to grab WALBufMappingLock to
1048 : * initialize it, but the WAL writer tries to do that ahead of insertions
1049 : * to avoid that from happening in the critical path.
1050 : *
1051 : *----------
1052 : */
1053 29389778 : START_CRIT_SECTION();
1054 29389778 : if (isLogSwitch)
1055 344 : WALInsertLockAcquireExclusive();
1056 : else
1057 29389434 : WALInsertLockAcquire();
1058 :
1059 : /*
1060 : * Check to see if my copy of RedoRecPtr is out of date. If so, may have
1061 : * to go back and have the caller recompute everything. This can only
1062 : * happen just after a checkpoint, so it's better to be slow in this case
1063 : * and fast otherwise.
1064 : *
1065 : * Also check to see if fullPageWrites or forcePageWrites was just turned
1066 : * on; if we weren't already doing full-page writes then go back and
1067 : * recompute.
1068 : *
1069 : * If we aren't doing full-page writes then RedoRecPtr doesn't actually
1070 : * affect the contents of the XLOG record, so we'll update our local copy
1071 : * but not force a recomputation. (If doPageWrites was just turned off,
1072 : * we could recompute the record without full pages, but we choose not to
1073 : * bother.)
1074 : */
1075 29389778 : if (RedoRecPtr != Insert->RedoRecPtr)
1076 : {
1077 : Assert(RedoRecPtr < Insert->RedoRecPtr);
1078 728 : RedoRecPtr = Insert->RedoRecPtr;
1079 : }
1080 29389778 : doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
1081 :
1082 29389778 : if (doPageWrites &&
1083 29233508 : (!prevDoPageWrites ||
1084 28206486 : (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
1085 : {
1086 : /*
1087 : * Oops, some buffer now needs to be backed up that the caller didn't
1088 : * back up. Start over.
1089 : */
1090 30 : WALInsertLockRelease();
1091 30 : END_CRIT_SECTION();
1092 30 : return InvalidXLogRecPtr;
1093 : }
1094 :
1095 : /*
1096 : * Reserve space for the record in the WAL. This also sets the xl_prev
1097 : * pointer.
1098 : */
1099 29389748 : if (isLogSwitch)
1100 344 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
1101 : else
1102 : {
1103 29389404 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
1104 : &rechdr->xl_prev);
1105 29389404 : inserted = true;
1106 : }
1107 :
1108 29389748 : if (inserted)
1109 : {
1110 : /*
1111 : * Now that xl_prev has been filled in, calculate CRC of the record
1112 : * header.
1113 : */
1114 29389694 : rdata_crc = rechdr->xl_crc;
1115 29389694 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
1116 29389694 : FIN_CRC32C(rdata_crc);
1117 29389694 : rechdr->xl_crc = rdata_crc;
1118 :
1119 : /*
1120 : * All the record data, including the header, is now ready to be
1121 : * inserted. Copy the record in the space reserved.
1122 : */
1123 29389694 : CopyXLogRecordToWAL(rechdr->xl_tot_len, isLogSwitch, rdata,
1124 : StartPos, EndPos);
1125 :
1126 : /*
1127 : * Unless record is flagged as not important, update LSN of last
1128 : * important record in the current slot. When holding all locks, just
1129 : * update the first one.
1130 : */
1131 29389694 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
1132 : {
1133 29241236 : int lockno = holdingAllLocks ? 0 : MyLockNo;
1134 :
1135 29241236 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
1136 : }
1137 : }
1138 : else
1139 : {
1140 : /*
1141 : * This was an xlog-switch record, but the current insert location was
1142 : * already exactly at the beginning of a segment, so there was no need
1143 : * to do anything.
1144 : */
1145 : }
1146 :
1147 : /*
1148 : * Done! Let others know that we're finished.
1149 : */
1150 29389748 : WALInsertLockRelease();
1151 :
1152 29389748 : MarkCurrentTransactionIdLoggedIfAny();
1153 :
1154 29389748 : END_CRIT_SECTION();
1155 :
1156 : /*
1157 : * Update shared LogwrtRqst.Write, if we crossed page boundary.
1158 : */
1159 29389748 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
1160 : {
1161 469472 : SpinLockAcquire(&XLogCtl->info_lck);
1162 : /* advance global request to include new block(s) */
1163 469472 : if (XLogCtl->LogwrtRqst.Write < EndPos)
1164 469408 : XLogCtl->LogwrtRqst.Write = EndPos;
1165 : /* update local result copy while I have the chance */
1166 469472 : LogwrtResult = XLogCtl->LogwrtResult;
1167 469472 : SpinLockRelease(&XLogCtl->info_lck);
1168 : }
1169 :
1170 : /*
1171 : * If this was an XLOG_SWITCH record, flush the record and the empty
1172 : * padding space that fills the rest of the segment, and perform
1173 : * end-of-segment actions (eg, notifying archiver).
1174 : */
1175 29389748 : if (isLogSwitch)
1176 : {
1177 : TRACE_POSTGRESQL_WAL_SWITCH();
1178 344 : XLogFlush(EndPos);
1179 :
1180 : /*
1181 : * Even though we reserved the rest of the segment for us, which is
1182 : * reflected in EndPos, we return a pointer to just the end of the
1183 : * xlog-switch record.
1184 : */
1185 344 : if (inserted)
1186 : {
1187 290 : EndPos = StartPos + SizeOfXLogRecord;
1188 290 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
1189 : {
1190 2 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
1191 :
1192 2 : if (offset == EndPos % XLOG_BLCKSZ)
1193 0 : EndPos += SizeOfXLogLongPHD;
1194 : else
1195 2 : EndPos += SizeOfXLogShortPHD;
1196 : }
1197 : }
1198 : }
1199 :
1200 : #ifdef WAL_DEBUG
1201 : if (XLOG_DEBUG)
1202 : {
1203 : static XLogReaderState *debug_reader = NULL;
1204 : StringInfoData buf;
1205 : StringInfoData recordBuf;
1206 : char *errormsg = NULL;
1207 : MemoryContext oldCxt;
1208 :
1209 : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1210 :
1211 : initStringInfo(&buf);
1212 : appendStringInfo(&buf, "INSERT @ %X/%X: ",
1213 : (uint32) (EndPos >> 32), (uint32) EndPos);
1214 :
1215 : /*
1216 : * We have to piece together the WAL record data from the XLogRecData
1217 : * entries, so that we can pass it to the rm_desc function as one
1218 : * contiguous chunk.
1219 : */
1220 : initStringInfo(&recordBuf);
1221 : for (; rdata != NULL; rdata = rdata->next)
1222 : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1223 :
1224 : if (!debug_reader)
1225 : debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1226 : XL_ROUTINE(), NULL);
1227 :
1228 : if (!debug_reader)
1229 : {
1230 : appendStringInfoString(&buf, "error decoding record: out of memory");
1231 : }
1232 : else if (!DecodeXLogRecord(debug_reader, (XLogRecord *) recordBuf.data,
1233 : &errormsg))
1234 : {
1235 : appendStringInfo(&buf, "error decoding record: %s",
1236 : errormsg ? errormsg : "no error message");
1237 : }
1238 : else
1239 : {
1240 : appendStringInfoString(&buf, " - ");
1241 : xlog_outdesc(&buf, debug_reader);
1242 : }
1243 : elog(LOG, "%s", buf.data);
1244 :
1245 : pfree(buf.data);
1246 : pfree(recordBuf.data);
1247 : MemoryContextSwitchTo(oldCxt);
1248 : }
1249 : #endif
1250 :
1251 : /*
1252 : * Update our global variables
1253 : */
1254 29389748 : ProcLastRecPtr = StartPos;
1255 29389748 : XactLastRecEnd = EndPos;
1256 :
1257 : /* Report WAL traffic to the instrumentation. */
1258 29389748 : if (inserted)
1259 : {
1260 29389694 : pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1261 29389694 : pgWalUsage.wal_records++;
1262 29389694 : pgWalUsage.wal_fpi += num_fpi;
1263 : }
1264 :
1265 29389748 : return EndPos;
1266 : }
1267 :
1268 : /*
1269 : * Reserves the right amount of space for a record of given size from the WAL.
1270 : * *StartPos is set to the beginning of the reserved section, *EndPos to
1271 : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1272 : * used to set the xl_prev of this record.
1273 : *
1274 : * This is the performance critical part of XLogInsert that must be serialized
1275 : * across backends. The rest can happen mostly in parallel. Try to keep this
1276 : * section as short as possible, insertpos_lck can be heavily contended on a
1277 : * busy system.
1278 : *
1279 : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1280 : * where we actually copy the record to the reserved space.
1281 : */
1282 : static void
1283 29389404 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1284 : XLogRecPtr *PrevPtr)
1285 : {
1286 29389404 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1287 : uint64 startbytepos;
1288 : uint64 endbytepos;
1289 : uint64 prevbytepos;
1290 :
1291 29389404 : size = MAXALIGN(size);
1292 :
1293 : /* All (non xlog-switch) records should contain data. */
1294 : Assert(size > SizeOfXLogRecord);
1295 :
1296 : /*
1297 : * The duration the spinlock needs to be held is minimized by minimizing
1298 : * the calculations that have to be done while holding the lock. The
1299 : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1300 : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1301 : * page headers. The mapping between "usable" byte positions and physical
1302 : * positions (XLogRecPtrs) can be done outside the locked region, and
1303 : * because the usable byte position doesn't include any headers, reserving
1304 : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1305 : */
1306 29389404 : SpinLockAcquire(&Insert->insertpos_lck);
1307 :
1308 29389404 : startbytepos = Insert->CurrBytePos;
1309 29389404 : endbytepos = startbytepos + size;
1310 29389404 : prevbytepos = Insert->PrevBytePos;
1311 29389404 : Insert->CurrBytePos = endbytepos;
1312 29389404 : Insert->PrevBytePos = startbytepos;
1313 :
1314 29389404 : SpinLockRelease(&Insert->insertpos_lck);
1315 :
1316 29389404 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1317 29389404 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1318 29389404 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1319 :
1320 : /*
1321 : * Check that the conversions between "usable byte positions" and
1322 : * XLogRecPtrs work consistently in both directions.
1323 : */
1324 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1325 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1326 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1327 29389404 : }
1328 :
1329 : /*
1330 : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1331 : *
1332 : * A log-switch record is handled slightly differently. The rest of the
1333 : * segment will be reserved for this insertion, as indicated by the returned
1334 : * *EndPos value. However, if we are already at the beginning of the current
1335 : * segment, *StartPos and *EndPos are set to the current location without
1336 : * reserving any space, and the function returns false.
1337 : */
1338 : static bool
1339 344 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1340 : {
1341 344 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1342 : uint64 startbytepos;
1343 : uint64 endbytepos;
1344 : uint64 prevbytepos;
1345 344 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1346 : XLogRecPtr ptr;
1347 : uint32 segleft;
1348 :
1349 : /*
1350 : * These calculations are a bit heavy-weight to be done while holding a
1351 : * spinlock, but since we're holding all the WAL insertion locks, there
1352 : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1353 : * compete for it, but that's not called very frequently.
1354 : */
1355 344 : SpinLockAcquire(&Insert->insertpos_lck);
1356 :
1357 344 : startbytepos = Insert->CurrBytePos;
1358 :
1359 344 : ptr = XLogBytePosToEndRecPtr(startbytepos);
1360 344 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1361 : {
1362 54 : SpinLockRelease(&Insert->insertpos_lck);
1363 54 : *EndPos = *StartPos = ptr;
1364 54 : return false;
1365 : }
1366 :
1367 290 : endbytepos = startbytepos + size;
1368 290 : prevbytepos = Insert->PrevBytePos;
1369 :
1370 290 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1371 290 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1372 :
1373 290 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1374 290 : if (segleft != wal_segment_size)
1375 : {
1376 : /* consume the rest of the segment */
1377 290 : *EndPos += segleft;
1378 290 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1379 : }
1380 290 : Insert->CurrBytePos = endbytepos;
1381 290 : Insert->PrevBytePos = startbytepos;
1382 :
1383 290 : SpinLockRelease(&Insert->insertpos_lck);
1384 :
1385 290 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1386 :
1387 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
1388 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1389 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1390 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1391 :
1392 290 : return true;
1393 : }
1394 :
1395 : /*
1396 : * Checks whether the current buffer page and backup page stored in the
1397 : * WAL record are consistent or not. Before comparing the two pages, a
1398 : * masking can be applied to the pages to ignore certain areas like hint bits,
1399 : * unused space between pd_lower and pd_upper among other things. This
1400 : * function should be called once WAL replay has been completed for a
1401 : * given record.
1402 : */
1403 : static void
1404 0 : checkXLogConsistency(XLogReaderState *record)
1405 : {
1406 0 : RmgrId rmid = XLogRecGetRmid(record);
1407 : RelFileNode rnode;
1408 : ForkNumber forknum;
1409 : BlockNumber blkno;
1410 : int block_id;
1411 :
1412 : /* Records with no backup blocks have no need for consistency checks. */
1413 0 : if (!XLogRecHasAnyBlockRefs(record))
1414 0 : return;
1415 :
1416 : Assert((XLogRecGetInfo(record) & XLR_CHECK_CONSISTENCY) != 0);
1417 :
1418 0 : for (block_id = 0; block_id <= record->max_block_id; block_id++)
1419 : {
1420 : Buffer buf;
1421 : Page page;
1422 :
1423 0 : if (!XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blkno))
1424 : {
1425 : /*
1426 : * WAL record doesn't contain a block reference with the given id.
1427 : * Do nothing.
1428 : */
1429 0 : continue;
1430 : }
1431 :
1432 : Assert(XLogRecHasBlockImage(record, block_id));
1433 :
1434 0 : if (XLogRecBlockImageApply(record, block_id))
1435 : {
1436 : /*
1437 : * WAL record has already applied the page, so bypass the
1438 : * consistency check as that would result in comparing the full
1439 : * page stored in the record with itself.
1440 : */
1441 0 : continue;
1442 : }
1443 :
1444 : /*
1445 : * Read the contents from the current buffer and store it in a
1446 : * temporary page.
1447 : */
1448 0 : buf = XLogReadBufferExtended(rnode, forknum, blkno,
1449 : RBM_NORMAL_NO_LOG);
1450 0 : if (!BufferIsValid(buf))
1451 0 : continue;
1452 :
1453 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
1454 0 : page = BufferGetPage(buf);
1455 :
1456 : /*
1457 : * Take a copy of the local page where WAL has been applied to have a
1458 : * comparison base before masking it...
1459 : */
1460 0 : memcpy(replay_image_masked, page, BLCKSZ);
1461 :
1462 : /* No need for this page anymore now that a copy is in. */
1463 0 : UnlockReleaseBuffer(buf);
1464 :
1465 : /*
1466 : * If the block LSN is already ahead of this WAL record, we can't
1467 : * expect contents to match. This can happen if recovery is
1468 : * restarted.
1469 : */
1470 0 : if (PageGetLSN(replay_image_masked) > record->EndRecPtr)
1471 0 : continue;
1472 :
1473 : /*
1474 : * Read the contents from the backup copy, stored in WAL record and
1475 : * store it in a temporary page. There is no need to allocate a new
1476 : * page here, a local buffer is fine to hold its contents and a mask
1477 : * can be directly applied on it.
1478 : */
1479 0 : if (!RestoreBlockImage(record, block_id, master_image_masked))
1480 0 : elog(ERROR, "failed to restore block image");
1481 :
1482 : /*
1483 : * If masking function is defined, mask both the master and replay
1484 : * images
1485 : */
1486 0 : if (RmgrTable[rmid].rm_mask != NULL)
1487 : {
1488 0 : RmgrTable[rmid].rm_mask(replay_image_masked, blkno);
1489 0 : RmgrTable[rmid].rm_mask(master_image_masked, blkno);
1490 : }
1491 :
1492 : /* Time to compare the master and replay images. */
1493 0 : if (memcmp(replay_image_masked, master_image_masked, BLCKSZ) != 0)
1494 : {
1495 0 : elog(FATAL,
1496 : "inconsistent page found, rel %u/%u/%u, forknum %u, blkno %u",
1497 : rnode.spcNode, rnode.dbNode, rnode.relNode,
1498 : forknum, blkno);
1499 : }
1500 : }
1501 : }
1502 :
1503 : /*
1504 : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1505 : * area in the WAL.
1506 : */
1507 : static void
1508 29389694 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1509 : XLogRecPtr StartPos, XLogRecPtr EndPos)
1510 : {
1511 : char *currpos;
1512 : int freespace;
1513 : int written;
1514 : XLogRecPtr CurrPos;
1515 : XLogPageHeader pagehdr;
1516 :
1517 : /*
1518 : * Get a pointer to the right place in the right WAL buffer to start
1519 : * inserting to.
1520 : */
1521 29389694 : CurrPos = StartPos;
1522 29389694 : currpos = GetXLogBuffer(CurrPos);
1523 29389694 : freespace = INSERT_FREESPACE(CurrPos);
1524 :
1525 : /*
1526 : * there should be enough space for at least the first field (xl_tot_len)
1527 : * on this page.
1528 : */
1529 : Assert(freespace >= sizeof(uint32));
1530 :
1531 : /* Copy record data */
1532 29389694 : written = 0;
1533 131594392 : while (rdata != NULL)
1534 : {
1535 102204698 : char *rdata_data = rdata->data;
1536 102204698 : int rdata_len = rdata->len;
1537 :
1538 102668282 : while (rdata_len > freespace)
1539 : {
1540 : /*
1541 : * Write what fits on this page, and continue on the next page.
1542 : */
1543 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1544 463584 : memcpy(currpos, rdata_data, freespace);
1545 463584 : rdata_data += freespace;
1546 463584 : rdata_len -= freespace;
1547 463584 : written += freespace;
1548 463584 : CurrPos += freespace;
1549 :
1550 : /*
1551 : * Get pointer to beginning of next page, and set the xlp_rem_len
1552 : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1553 : *
1554 : * It's safe to set the contrecord flag and xlp_rem_len without a
1555 : * lock on the page. All the other flags were already set when the
1556 : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1557 : * only backend that needs to set the contrecord flag.
1558 : */
1559 463584 : currpos = GetXLogBuffer(CurrPos);
1560 463584 : pagehdr = (XLogPageHeader) currpos;
1561 463584 : pagehdr->xlp_rem_len = write_len - written;
1562 463584 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1563 :
1564 : /* skip over the page header */
1565 463584 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1566 : {
1567 566 : CurrPos += SizeOfXLogLongPHD;
1568 566 : currpos += SizeOfXLogLongPHD;
1569 : }
1570 : else
1571 : {
1572 463018 : CurrPos += SizeOfXLogShortPHD;
1573 463018 : currpos += SizeOfXLogShortPHD;
1574 : }
1575 463584 : freespace = INSERT_FREESPACE(CurrPos);
1576 : }
1577 :
1578 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1579 102204698 : memcpy(currpos, rdata_data, rdata_len);
1580 102204698 : currpos += rdata_len;
1581 102204698 : CurrPos += rdata_len;
1582 102204698 : freespace -= rdata_len;
1583 102204698 : written += rdata_len;
1584 :
1585 102204698 : rdata = rdata->next;
1586 : }
1587 : Assert(written == write_len);
1588 :
1589 : /*
1590 : * If this was an xlog-switch, it's not enough to write the switch record,
1591 : * we also have to consume all the remaining space in the WAL segment. We
1592 : * have already reserved that space, but we need to actually fill it.
1593 : */
1594 29389984 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1595 : {
1596 : /* An xlog-switch record doesn't contain any data besides the header */
1597 : Assert(write_len == SizeOfXLogRecord);
1598 :
1599 : /* Assert that we did reserve the right amount of space */
1600 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1601 :
1602 : /* Use up all the remaining space on the current page */
1603 290 : CurrPos += freespace;
1604 :
1605 : /*
1606 : * Cause all remaining pages in the segment to be flushed, leaving the
1607 : * XLog position where it should be, at the start of the next segment.
1608 : * We do this one page at a time, to make sure we don't deadlock
1609 : * against ourselves if wal_buffers < wal_segment_size.
1610 : */
1611 434484 : while (CurrPos < EndPos)
1612 : {
1613 : /*
1614 : * The minimal action to flush the page would be to call
1615 : * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1616 : * AdvanceXLInsertBuffer(...). The page would be left initialized
1617 : * mostly to zeros, except for the page header (always the short
1618 : * variant, as this is never a segment's first page).
1619 : *
1620 : * The large vistas of zeros are good for compressibility, but the
1621 : * headers interrupting them every XLOG_BLCKSZ (with values that
1622 : * differ from page to page) are not. The effect varies with
1623 : * compression tool, but bzip2 for instance compresses about an
1624 : * order of magnitude worse if those headers are left in place.
1625 : *
1626 : * Rather than complicating AdvanceXLInsertBuffer itself (which is
1627 : * called in heavily-loaded circumstances as well as this lightly-
1628 : * loaded one) with variant behavior, we just use GetXLogBuffer
1629 : * (which itself calls the two methods we need) to get the pointer
1630 : * and zero most of the page. Then we just zero the page header.
1631 : */
1632 434194 : currpos = GetXLogBuffer(CurrPos);
1633 1736776 : MemSet(currpos, 0, SizeOfXLogShortPHD);
1634 :
1635 434194 : CurrPos += XLOG_BLCKSZ;
1636 : }
1637 : }
1638 : else
1639 : {
1640 : /* Align the end position, so that the next record starts aligned */
1641 29389404 : CurrPos = MAXALIGN64(CurrPos);
1642 : }
1643 :
1644 29389694 : if (CurrPos != EndPos)
1645 0 : elog(PANIC, "space reserved for WAL record does not match what was written");
1646 29389694 : }
1647 :
1648 : /*
1649 : * Acquire a WAL insertion lock, for inserting to WAL.
1650 : */
1651 : static void
1652 29389434 : WALInsertLockAcquire(void)
1653 : {
1654 : bool immed;
1655 :
1656 : /*
1657 : * It doesn't matter which of the WAL insertion locks we acquire, so try
1658 : * the one we used last time. If the system isn't particularly busy, it's
1659 : * a good bet that it's still available, and it's good to have some
1660 : * affinity to a particular lock so that you don't unnecessarily bounce
1661 : * cache lines between processes when there's no contention.
1662 : *
1663 : * If this is the first time through in this backend, pick a lock
1664 : * (semi-)randomly. This allows the locks to be used evenly if you have a
1665 : * lot of very short connections.
1666 : */
1667 : static int lockToTry = -1;
1668 :
1669 29389434 : if (lockToTry == -1)
1670 6800 : lockToTry = MyProc->pgprocno % NUM_XLOGINSERT_LOCKS;
1671 29389434 : MyLockNo = lockToTry;
1672 :
1673 : /*
1674 : * The insertingAt value is initially set to 0, as we don't know our
1675 : * insert location yet.
1676 : */
1677 29389434 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
1678 29389434 : if (!immed)
1679 : {
1680 : /*
1681 : * If we couldn't get the lock immediately, try another lock next
1682 : * time. On a system with more insertion locks than concurrent
1683 : * inserters, this causes all the inserters to eventually migrate to a
1684 : * lock that no-one else is using. On a system with more inserters
1685 : * than locks, it still helps to distribute the inserters evenly
1686 : * across the locks.
1687 : */
1688 228 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1689 : }
1690 29389434 : }
1691 :
1692 : /*
1693 : * Acquire all WAL insertion locks, to prevent other backends from inserting
1694 : * to WAL.
1695 : */
1696 : static void
1697 4062 : WALInsertLockAcquireExclusive(void)
1698 : {
1699 : int i;
1700 :
1701 : /*
1702 : * When holding all the locks, all but the last lock's insertingAt
1703 : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1704 : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1705 : */
1706 32496 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1707 : {
1708 28434 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1709 28434 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1710 28434 : &WALInsertLocks[i].l.insertingAt,
1711 : PG_UINT64_MAX);
1712 : }
1713 : /* Variable value reset to 0 at release */
1714 4062 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1715 :
1716 4062 : holdingAllLocks = true;
1717 4062 : }
1718 :
1719 : /*
1720 : * Release our insertion lock (or locks, if we're holding them all).
1721 : *
1722 : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1723 : * next time the lock is acquired.
1724 : */
1725 : static void
1726 29393496 : WALInsertLockRelease(void)
1727 : {
1728 29393496 : if (holdingAllLocks)
1729 : {
1730 : int i;
1731 :
1732 36558 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1733 32496 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1734 32496 : &WALInsertLocks[i].l.insertingAt,
1735 : 0);
1736 :
1737 4062 : holdingAllLocks = false;
1738 : }
1739 : else
1740 : {
1741 29389434 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1742 29389434 : &WALInsertLocks[MyLockNo].l.insertingAt,
1743 : 0);
1744 : }
1745 29393496 : }
1746 :
1747 : /*
1748 : * Update our insertingAt value, to let others know that we've finished
1749 : * inserting up to that point.
1750 : */
1751 : static void
1752 729194 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1753 : {
1754 729194 : if (holdingAllLocks)
1755 : {
1756 : /*
1757 : * We use the last lock to mark our actual position, see comments in
1758 : * WALInsertLockAcquireExclusive.
1759 : */
1760 429054 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
1761 429054 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1762 : insertingAt);
1763 : }
1764 : else
1765 300140 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1766 300140 : &WALInsertLocks[MyLockNo].l.insertingAt,
1767 : insertingAt);
1768 729194 : }
1769 :
1770 : /*
1771 : * Wait for any WAL insertions < upto to finish.
1772 : *
1773 : * Returns the location of the oldest insertion that is still in-progress.
1774 : * Any WAL prior to that point has been fully copied into WAL buffers, and
1775 : * can be flushed out to disk. Because this waits for any insertions older
1776 : * than 'upto' to finish, the return value is always >= 'upto'.
1777 : *
1778 : * Note: When you are about to write out WAL, you must call this function
1779 : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1780 : * need to wait for an insertion to finish (or at least advance to next
1781 : * uninitialized page), and the inserter might need to evict an old WAL buffer
1782 : * to make room for a new one, which in turn requires WALWriteLock.
1783 : */
1784 : static XLogRecPtr
1785 692660 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1786 : {
1787 : uint64 bytepos;
1788 : XLogRecPtr reservedUpto;
1789 : XLogRecPtr finishedUpto;
1790 692660 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1791 : int i;
1792 :
1793 692660 : if (MyProc == NULL)
1794 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1795 :
1796 : /* Read the current insert position */
1797 692660 : SpinLockAcquire(&Insert->insertpos_lck);
1798 692660 : bytepos = Insert->CurrBytePos;
1799 692660 : SpinLockRelease(&Insert->insertpos_lck);
1800 692660 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1801 :
1802 : /*
1803 : * No-one should request to flush a piece of WAL that hasn't even been
1804 : * reserved yet. However, it can happen if there is a block with a bogus
1805 : * LSN on disk, for example. XLogFlush checks for that situation and
1806 : * complains, but only after the flush. Here we just assume that to mean
1807 : * that all WAL that has been reserved needs to be finished. In this
1808 : * corner-case, the return value can be smaller than 'upto' argument.
1809 : */
1810 692660 : if (upto > reservedUpto)
1811 : {
1812 0 : elog(LOG, "request to flush past end of generated WAL; request %X/%X, currpos %X/%X",
1813 : (uint32) (upto >> 32), (uint32) upto,
1814 : (uint32) (reservedUpto >> 32), (uint32) reservedUpto);
1815 0 : upto = reservedUpto;
1816 : }
1817 :
1818 : /*
1819 : * Loop through all the locks, sleeping on any in-progress insert older
1820 : * than 'upto'.
1821 : *
1822 : * finishedUpto is our return value, indicating the point upto which all
1823 : * the WAL insertions have been finished. Initialize it to the head of
1824 : * reserved WAL, and as we iterate through the insertion locks, back it
1825 : * out for any insertion that's still in progress.
1826 : */
1827 692660 : finishedUpto = reservedUpto;
1828 6233940 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1829 : {
1830 5541280 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1831 :
1832 : do
1833 : {
1834 : /*
1835 : * See if this insertion is in progress. LWLockWaitForVar will
1836 : * wait for the lock to be released, or for the 'value' to be set
1837 : * by a LWLockUpdateVar call. When a lock is initially acquired,
1838 : * its value is 0 (InvalidXLogRecPtr), which means that we don't
1839 : * know where it's inserting yet. We will have to wait for it. If
1840 : * it's a small insertion, the record will most likely fit on the
1841 : * same page and the inserter will release the lock without ever
1842 : * calling LWLockUpdateVar. But if it has to sleep, it will
1843 : * advertise the insertion point with LWLockUpdateVar before
1844 : * sleeping.
1845 : */
1846 5541422 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1847 5541422 : &WALInsertLocks[i].l.insertingAt,
1848 : insertingat, &insertingat))
1849 : {
1850 : /* the lock was free, so no insertion in progress */
1851 2328508 : insertingat = InvalidXLogRecPtr;
1852 2328508 : break;
1853 : }
1854 :
1855 : /*
1856 : * This insertion is still in progress. Have to wait, unless the
1857 : * inserter has proceeded past 'upto'.
1858 : */
1859 3212914 : } while (insertingat < upto);
1860 :
1861 5541280 : if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1862 420146 : finishedUpto = insertingat;
1863 : }
1864 692660 : return finishedUpto;
1865 : }
1866 :
1867 : /*
1868 : * Get a pointer to the right location in the WAL buffer containing the
1869 : * given XLogRecPtr.
1870 : *
1871 : * If the page is not initialized yet, it is initialized. That might require
1872 : * evicting an old dirty buffer from the buffer cache, which means I/O.
1873 : *
1874 : * The caller must ensure that the page containing the requested location
1875 : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1876 : * hold onto a WAL insertion lock with the insertingAt position set to
1877 : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1878 : * to evict an old page from the buffer. (This means that once you call
1879 : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1880 : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1881 : * later, because older buffers might be recycled already)
1882 : */
1883 : static char *
1884 30287472 : GetXLogBuffer(XLogRecPtr ptr)
1885 : {
1886 : int idx;
1887 : XLogRecPtr endptr;
1888 : static uint64 cachedPage = 0;
1889 : static char *cachedPos = NULL;
1890 : XLogRecPtr expectedEndPtr;
1891 :
1892 : /*
1893 : * Fast path for the common case that we need to access again the same
1894 : * page as last time.
1895 : */
1896 30287472 : if (ptr / XLOG_BLCKSZ == cachedPage)
1897 : {
1898 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1899 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1900 29273928 : return cachedPos + ptr % XLOG_BLCKSZ;
1901 : }
1902 :
1903 : /*
1904 : * The XLog buffer cache is organized so that a page is always loaded to a
1905 : * particular buffer. That way we can easily calculate the buffer a given
1906 : * page must be loaded into, from the XLogRecPtr alone.
1907 : */
1908 1013544 : idx = XLogRecPtrToBufIdx(ptr);
1909 :
1910 : /*
1911 : * See what page is loaded in the buffer at the moment. It could be the
1912 : * page we're looking for, or something older. It can't be anything newer
1913 : * - that would imply the page we're looking for has already been written
1914 : * out to disk and evicted, and the caller is responsible for making sure
1915 : * that doesn't happen.
1916 : *
1917 : * However, we don't hold a lock while we read the value. If someone has
1918 : * just initialized the page, it's possible that we get a "torn read" of
1919 : * the XLogRecPtr if 64-bit fetches are not atomic on this platform. In
1920 : * that case we will see a bogus value. That's ok, we'll grab the mapping
1921 : * lock (in AdvanceXLInsertBuffer) and retry if we see anything else than
1922 : * the page we're looking for. But it means that when we do this unlocked
1923 : * read, we might see a value that appears to be ahead of the page we're
1924 : * looking for. Don't PANIC on that, until we've verified the value while
1925 : * holding the lock.
1926 : */
1927 1013544 : expectedEndPtr = ptr;
1928 1013544 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1929 :
1930 1013544 : endptr = XLogCtl->xlblocks[idx];
1931 1013544 : if (expectedEndPtr != endptr)
1932 : {
1933 : XLogRecPtr initializedUpto;
1934 :
1935 : /*
1936 : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1937 : * know how far we're finished with inserting the record.
1938 : *
1939 : * NB: If 'ptr' points to just after the page header, advertise a
1940 : * position at the beginning of the page rather than 'ptr' itself. If
1941 : * there are no other insertions running, someone might try to flush
1942 : * up to our advertised location. If we advertised a position after
1943 : * the page header, someone might try to flush the page header, even
1944 : * though page might actually not be initialized yet. As the first
1945 : * inserter on the page, we are effectively responsible for making
1946 : * sure that it's initialized, before we let insertingAt to move past
1947 : * the page header.
1948 : */
1949 729194 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
1950 14146 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
1951 14146 : initializedUpto = ptr - SizeOfXLogShortPHD;
1952 715048 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
1953 240 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
1954 190 : initializedUpto = ptr - SizeOfXLogLongPHD;
1955 : else
1956 714858 : initializedUpto = ptr;
1957 :
1958 729194 : WALInsertLockUpdateInsertingAt(initializedUpto);
1959 :
1960 729194 : AdvanceXLInsertBuffer(ptr, false);
1961 729194 : endptr = XLogCtl->xlblocks[idx];
1962 :
1963 729194 : if (expectedEndPtr != endptr)
1964 0 : elog(PANIC, "could not find WAL buffer for %X/%X",
1965 : (uint32) (ptr >> 32), (uint32) ptr);
1966 : }
1967 : else
1968 : {
1969 : /*
1970 : * Make sure the initialization of the page is visible to us, and
1971 : * won't arrive later to overwrite the WAL data we write on the page.
1972 : */
1973 284350 : pg_memory_barrier();
1974 : }
1975 :
1976 : /*
1977 : * Found the buffer holding this page. Return a pointer to the right
1978 : * offset within the page.
1979 : */
1980 1013544 : cachedPage = ptr / XLOG_BLCKSZ;
1981 1013544 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1982 :
1983 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1984 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1985 :
1986 1013544 : return cachedPos + ptr % XLOG_BLCKSZ;
1987 : }
1988 :
1989 : /*
1990 : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1991 : * is the position starting from the beginning of WAL, excluding all WAL
1992 : * page headers.
1993 : */
1994 : static XLogRecPtr
1995 58806640 : XLogBytePosToRecPtr(uint64 bytepos)
1996 : {
1997 : uint64 fullsegs;
1998 : uint64 fullpages;
1999 : uint64 bytesleft;
2000 : uint32 seg_offset;
2001 : XLogRecPtr result;
2002 :
2003 58806640 : fullsegs = bytepos / UsableBytesInSegment;
2004 58806640 : bytesleft = bytepos % UsableBytesInSegment;
2005 :
2006 58806640 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
2007 : {
2008 : /* fits on first page of segment */
2009 101902 : seg_offset = bytesleft + SizeOfXLogLongPHD;
2010 : }
2011 : else
2012 : {
2013 : /* account for the first page on segment with long header */
2014 58704738 : seg_offset = XLOG_BLCKSZ;
2015 58704738 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
2016 :
2017 58704738 : fullpages = bytesleft / UsableBytesInPage;
2018 58704738 : bytesleft = bytesleft % UsableBytesInPage;
2019 :
2020 58704738 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
2021 : }
2022 :
2023 58806640 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
2024 :
2025 58806640 : return result;
2026 : }
2027 :
2028 : /*
2029 : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
2030 : * returns a pointer to the beginning of the page (ie. before page header),
2031 : * not to where the first xlog record on that page would go to. This is used
2032 : * when converting a pointer to the end of a record.
2033 : */
2034 : static XLogRecPtr
2035 30082698 : XLogBytePosToEndRecPtr(uint64 bytepos)
2036 : {
2037 : uint64 fullsegs;
2038 : uint64 fullpages;
2039 : uint64 bytesleft;
2040 : uint32 seg_offset;
2041 : XLogRecPtr result;
2042 :
2043 30082698 : fullsegs = bytepos / UsableBytesInSegment;
2044 30082698 : bytesleft = bytepos % UsableBytesInSegment;
2045 :
2046 30082698 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
2047 : {
2048 : /* fits on first page of segment */
2049 450242 : if (bytesleft == 0)
2050 399290 : seg_offset = 0;
2051 : else
2052 50952 : seg_offset = bytesleft + SizeOfXLogLongPHD;
2053 : }
2054 : else
2055 : {
2056 : /* account for the first page on segment with long header */
2057 29632456 : seg_offset = XLOG_BLCKSZ;
2058 29632456 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
2059 :
2060 29632456 : fullpages = bytesleft / UsableBytesInPage;
2061 29632456 : bytesleft = bytesleft % UsableBytesInPage;
2062 :
2063 29632456 : if (bytesleft == 0)
2064 27786 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
2065 : else
2066 29604670 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
2067 : }
2068 :
2069 30082698 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
2070 :
2071 30082698 : return result;
2072 : }
2073 :
2074 : /*
2075 : * Convert an XLogRecPtr to a "usable byte position".
2076 : */
2077 : static uint64
2078 3002 : XLogRecPtrToBytePos(XLogRecPtr ptr)
2079 : {
2080 : uint64 fullsegs;
2081 : uint32 fullpages;
2082 : uint32 offset;
2083 : uint64 result;
2084 :
2085 3002 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
2086 :
2087 3002 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
2088 3002 : offset = ptr % XLOG_BLCKSZ;
2089 :
2090 3002 : if (fullpages == 0)
2091 : {
2092 1094 : result = fullsegs * UsableBytesInSegment;
2093 1094 : if (offset > 0)
2094 : {
2095 : Assert(offset >= SizeOfXLogLongPHD);
2096 786 : result += offset - SizeOfXLogLongPHD;
2097 : }
2098 : }
2099 : else
2100 : {
2101 3816 : result = fullsegs * UsableBytesInSegment +
2102 3816 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
2103 1908 : (fullpages - 1) * UsableBytesInPage; /* full pages */
2104 1908 : if (offset > 0)
2105 : {
2106 : Assert(offset >= SizeOfXLogShortPHD);
2107 1908 : result += offset - SizeOfXLogShortPHD;
2108 : }
2109 : }
2110 :
2111 3002 : return result;
2112 : }
2113 :
2114 : /*
2115 : * Initialize XLOG buffers, writing out old buffers if they still contain
2116 : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
2117 : * true, initialize as many pages as we can without having to write out
2118 : * unwritten data. Any new pages are initialized to zeros, with pages headers
2119 : * initialized properly.
2120 : */
2121 : static void
2122 738544 : AdvanceXLInsertBuffer(XLogRecPtr upto, bool opportunistic)
2123 : {
2124 738544 : XLogCtlInsert *Insert = &XLogCtl->Insert;
2125 : int nextidx;
2126 : XLogRecPtr OldPageRqstPtr;
2127 : XLogwrtRqst WriteRqst;
2128 738544 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
2129 : XLogRecPtr NewPageBeginPtr;
2130 : XLogPageHeader NewPage;
2131 738544 : int npages = 0;
2132 :
2133 738544 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2134 :
2135 : /*
2136 : * Now that we have the lock, check if someone initialized the page
2137 : * already.
2138 : */
2139 2120960 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2140 : {
2141 1391766 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2142 :
2143 : /*
2144 : * Get ending-offset of the buffer page we need to replace (this may
2145 : * be zero if the buffer hasn't been used yet). Fall through if it's
2146 : * already written out.
2147 : */
2148 1391766 : OldPageRqstPtr = XLogCtl->xlblocks[nextidx];
2149 1391766 : if (LogwrtResult.Write < OldPageRqstPtr)
2150 : {
2151 : /*
2152 : * Nope, got work to do. If we just want to pre-initialize as much
2153 : * as we can without flushing, give up now.
2154 : */
2155 429782 : if (opportunistic)
2156 9350 : break;
2157 :
2158 : /* Before waiting, get info_lck and update LogwrtResult */
2159 420432 : SpinLockAcquire(&XLogCtl->info_lck);
2160 420432 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2161 398900 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2162 420432 : LogwrtResult = XLogCtl->LogwrtResult;
2163 420432 : SpinLockRelease(&XLogCtl->info_lck);
2164 :
2165 : /*
2166 : * Now that we have an up-to-date LogwrtResult value, see if we
2167 : * still need to write it or if someone else already did.
2168 : */
2169 420432 : if (LogwrtResult.Write < OldPageRqstPtr)
2170 : {
2171 : /*
2172 : * Must acquire write lock. Release WALBufMappingLock first,
2173 : * to make sure that all insertions that we need to wait for
2174 : * can finish (up to this same position). Otherwise we risk
2175 : * deadlock.
2176 : */
2177 420066 : LWLockRelease(WALBufMappingLock);
2178 :
2179 420066 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2180 :
2181 420066 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2182 :
2183 420066 : LogwrtResult = XLogCtl->LogwrtResult;
2184 420066 : if (LogwrtResult.Write >= OldPageRqstPtr)
2185 : {
2186 : /* OK, someone wrote it already */
2187 72 : LWLockRelease(WALWriteLock);
2188 : }
2189 : else
2190 : {
2191 : /* Have to write it ourselves */
2192 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2193 419994 : WriteRqst.Write = OldPageRqstPtr;
2194 419994 : WriteRqst.Flush = 0;
2195 419994 : XLogWrite(WriteRqst, false);
2196 419994 : LWLockRelease(WALWriteLock);
2197 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2198 : }
2199 : /* Re-acquire WALBufMappingLock and retry */
2200 420066 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2201 420066 : continue;
2202 : }
2203 : }
2204 :
2205 : /*
2206 : * Now the next buffer slot is free and we can set it up to be the
2207 : * next output page.
2208 : */
2209 962350 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
2210 962350 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2211 :
2212 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2213 :
2214 962350 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2215 :
2216 : /*
2217 : * Be sure to re-zero the buffer so that bytes beyond what we've
2218 : * written will look like zeroes and not valid XLOG records...
2219 : */
2220 962350 : MemSet((char *) NewPage, 0, XLOG_BLCKSZ);
2221 :
2222 : /*
2223 : * Fill the new page's header
2224 : */
2225 962350 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2226 :
2227 : /* NewPage->xlp_info = 0; */ /* done by memset */
2228 962350 : NewPage->xlp_tli = ThisTimeLineID;
2229 962350 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2230 :
2231 : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2232 :
2233 : /*
2234 : * If online backup is not in progress, mark the header to indicate
2235 : * that WAL records beginning in this page have removable backup
2236 : * blocks. This allows the WAL archiver to know whether it is safe to
2237 : * compress archived WAL data by transforming full-block records into
2238 : * the non-full-block format. It is sufficient to record this at the
2239 : * page level because we force a page switch (in fact a segment
2240 : * switch) when starting a backup, so the flag will be off before any
2241 : * records can be written during the backup. At the end of a backup,
2242 : * the last page will be marked as all unsafe when perhaps only part
2243 : * is unsafe, but at worst the archiver would miss the opportunity to
2244 : * compress a few records.
2245 : */
2246 962350 : if (!Insert->forcePageWrites)
2247 831946 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2248 :
2249 : /*
2250 : * If first page of an XLOG segment file, make it a long header.
2251 : */
2252 962350 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2253 : {
2254 790 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2255 :
2256 790 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2257 790 : NewLongPage->xlp_seg_size = wal_segment_size;
2258 790 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
2259 790 : NewPage->xlp_info |= XLP_LONG_HEADER;
2260 : }
2261 :
2262 : /*
2263 : * Make sure the initialization of the page becomes visible to others
2264 : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2265 : * holding a lock.
2266 : */
2267 962350 : pg_write_barrier();
2268 :
2269 962350 : *((volatile XLogRecPtr *) &XLogCtl->xlblocks[nextidx]) = NewPageEndPtr;
2270 :
2271 962350 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2272 :
2273 962350 : npages++;
2274 : }
2275 738544 : LWLockRelease(WALBufMappingLock);
2276 :
2277 : #ifdef WAL_DEBUG
2278 : if (XLOG_DEBUG && npages > 0)
2279 : {
2280 : elog(DEBUG1, "initialized %d pages, up to %X/%X",
2281 : npages, (uint32) (NewPageEndPtr >> 32), (uint32) NewPageEndPtr);
2282 : }
2283 : #endif
2284 738544 : }
2285 :
2286 : /*
2287 : * Calculate CheckPointSegments based on max_wal_size_mb and
2288 : * checkpoint_completion_target.
2289 : */
2290 : static void
2291 8558 : CalculateCheckpointSegments(void)
2292 : {
2293 : double target;
2294 :
2295 : /*-------
2296 : * Calculate the distance at which to trigger a checkpoint, to avoid
2297 : * exceeding max_wal_size_mb. This is based on two assumptions:
2298 : *
2299 : * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2300 : * WAL for two checkpoint cycles to allow us to recover from the
2301 : * secondary checkpoint if the first checkpoint failed, though we
2302 : * only did this on the master anyway, not on standby. Keeping just
2303 : * one checkpoint simplifies processing and reduces disk space in
2304 : * many smaller databases.)
2305 : * b) during checkpoint, we consume checkpoint_completion_target *
2306 : * number of segments consumed between checkpoints.
2307 : *-------
2308 : */
2309 17116 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2310 8558 : (1.0 + CheckPointCompletionTarget);
2311 :
2312 : /* round down */
2313 8558 : CheckPointSegments = (int) target;
2314 :
2315 8558 : if (CheckPointSegments < 1)
2316 14 : CheckPointSegments = 1;
2317 8558 : }
2318 :
2319 : void
2320 7094 : assign_max_wal_size(int newval, void *extra)
2321 : {
2322 7094 : max_wal_size_mb = newval;
2323 7094 : CalculateCheckpointSegments();
2324 7094 : }
2325 :
2326 : void
2327 0 : assign_checkpoint_completion_target(double newval, void *extra)
2328 : {
2329 0 : CheckPointCompletionTarget = newval;
2330 0 : CalculateCheckpointSegments();
2331 0 : }
2332 :
2333 : /*
2334 : * At a checkpoint, how many WAL segments to recycle as preallocated future
2335 : * XLOG segments? Returns the highest segment that should be preallocated.
2336 : */
2337 : static XLogSegNo
2338 718 : XLOGfileslop(XLogRecPtr lastredoptr)
2339 : {
2340 : XLogSegNo minSegNo;
2341 : XLogSegNo maxSegNo;
2342 : double distance;
2343 : XLogSegNo recycleSegNo;
2344 :
2345 : /*
2346 : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2347 : * correspond to. Always recycle enough segments to meet the minimum, and
2348 : * remove enough segments to stay below the maximum.
2349 : */
2350 1436 : minSegNo = lastredoptr / wal_segment_size +
2351 718 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2352 1436 : maxSegNo = lastredoptr / wal_segment_size +
2353 718 : ConvertToXSegs(max_wal_size_mb, wal_segment_size) - 1;
2354 :
2355 : /*
2356 : * Between those limits, recycle enough segments to get us through to the
2357 : * estimated end of next checkpoint.
2358 : *
2359 : * To estimate where the next checkpoint will finish, assume that the
2360 : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2361 : * every checkpoint.
2362 : */
2363 718 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2364 : /* add 10% for good measure. */
2365 718 : distance *= 1.10;
2366 :
2367 718 : recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2368 : wal_segment_size);
2369 :
2370 718 : if (recycleSegNo < minSegNo)
2371 60 : recycleSegNo = minSegNo;
2372 718 : if (recycleSegNo > maxSegNo)
2373 514 : recycleSegNo = maxSegNo;
2374 :
2375 718 : return recycleSegNo;
2376 : }
2377 :
2378 : /*
2379 : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2380 : *
2381 : * new_segno indicates a log file that has just been filled up (or read
2382 : * during recovery). We measure the distance from RedoRecPtr to new_segno
2383 : * and see if that exceeds CheckPointSegments.
2384 : *
2385 : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2386 : */
2387 : static bool
2388 1070 : XLogCheckpointNeeded(XLogSegNo new_segno)
2389 : {
2390 : XLogSegNo old_segno;
2391 :
2392 1070 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2393 :
2394 1070 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2395 184 : return true;
2396 886 : return false;
2397 : }
2398 :
2399 : /*
2400 : * Write and/or fsync the log at least as far as WriteRqst indicates.
2401 : *
2402 : * If flexible == true, we don't have to write as far as WriteRqst, but
2403 : * may stop at any convenient boundary (such as a cache or logfile boundary).
2404 : * This option allows us to avoid uselessly issuing multiple writes when a
2405 : * single one would do.
2406 : *
2407 : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2408 : * must be called before grabbing the lock, to make sure the data is ready to
2409 : * write.
2410 : */
2411 : static void
2412 687966 : XLogWrite(XLogwrtRqst WriteRqst, bool flexible)
2413 : {
2414 : bool ispartialpage;
2415 : bool last_iteration;
2416 : bool finishing_seg;
2417 : bool use_existent;
2418 : int curridx;
2419 : int npages;
2420 : int startidx;
2421 : uint32 startoffset;
2422 :
2423 : /* We should always be inside a critical section here */
2424 : Assert(CritSectionCount > 0);
2425 :
2426 : /*
2427 : * Update local LogwrtResult (caller probably did this already, but...)
2428 : */
2429 687966 : LogwrtResult = XLogCtl->LogwrtResult;
2430 :
2431 : /*
2432 : * Since successive pages in the xlog cache are consecutively allocated,
2433 : * we can usually gather multiple pages together and issue just one
2434 : * write() call. npages is the number of pages we have determined can be
2435 : * written together; startidx is the cache block index of the first one,
2436 : * and startoffset is the file offset at which it should go. The latter
2437 : * two variables are only valid when npages > 0, but we must initialize
2438 : * all of them to keep the compiler quiet.
2439 : */
2440 687966 : npages = 0;
2441 687966 : startidx = 0;
2442 687966 : startoffset = 0;
2443 :
2444 : /*
2445 : * Within the loop, curridx is the cache block index of the page to
2446 : * consider writing. Begin at the buffer containing the next unwritten
2447 : * page, or last partially written page.
2448 : */
2449 687966 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2450 :
2451 1607448 : while (LogwrtResult.Write < WriteRqst.Write)
2452 : {
2453 : /*
2454 : * Make sure we're not ahead of the insert process. This could happen
2455 : * if we're passed a bogus WriteRqst.Write that is past the end of the
2456 : * last page that's been initialized by AdvanceXLInsertBuffer.
2457 : */
2458 1187012 : XLogRecPtr EndPtr = XLogCtl->xlblocks[curridx];
2459 :
2460 1187012 : if (LogwrtResult.Write >= EndPtr)
2461 0 : elog(PANIC, "xlog write request %X/%X is past end of log %X/%X",
2462 : (uint32) (LogwrtResult.Write >> 32),
2463 : (uint32) LogwrtResult.Write,
2464 : (uint32) (EndPtr >> 32), (uint32) EndPtr);
2465 :
2466 : /* Advance LogwrtResult.Write to end of current buffer page */
2467 1187012 : LogwrtResult.Write = EndPtr;
2468 1187012 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2469 :
2470 1187012 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2471 : wal_segment_size))
2472 : {
2473 : /*
2474 : * Switch to new logfile segment. We cannot have any pending
2475 : * pages here (since we dump what we have at segment end).
2476 : */
2477 : Assert(npages == 0);
2478 8230 : if (openLogFile >= 0)
2479 1944 : XLogFileClose();
2480 8230 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2481 : wal_segment_size);
2482 :
2483 : /* create/use new log file */
2484 8230 : use_existent = true;
2485 8230 : openLogFile = XLogFileInit(openLogSegNo, &use_existent, true);
2486 8230 : ReserveExternalFD();
2487 : }
2488 :
2489 : /* Make sure we have the current logfile open */
2490 1187012 : if (openLogFile < 0)
2491 : {
2492 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2493 : wal_segment_size);
2494 0 : openLogFile = XLogFileOpen(openLogSegNo);
2495 0 : ReserveExternalFD();
2496 : }
2497 :
2498 : /* Add current page to the set of pending pages-to-dump */
2499 1187012 : if (npages == 0)
2500 : {
2501 : /* first of group */
2502 689256 : startidx = curridx;
2503 689256 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2504 : wal_segment_size);
2505 : }
2506 1187012 : npages++;
2507 :
2508 : /*
2509 : * Dump the set if this will be the last loop iteration, or if we are
2510 : * at the last page of the cache area (since the next page won't be
2511 : * contiguous in memory), or if we are at the end of the logfile
2512 : * segment.
2513 : */
2514 1187012 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2515 :
2516 2111472 : finishing_seg = !ispartialpage &&
2517 924460 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2518 :
2519 1187012 : if (last_iteration ||
2520 499264 : curridx == XLogCtl->XLogCacheBlck ||
2521 : finishing_seg)
2522 : {
2523 : char *from;
2524 : Size nbytes;
2525 : Size nleft;
2526 : int written;
2527 :
2528 : /* OK to write the page(s) */
2529 689256 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2530 689256 : nbytes = npages * (Size) XLOG_BLCKSZ;
2531 689256 : nleft = nbytes;
2532 : do
2533 : {
2534 689256 : errno = 0;
2535 689256 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
2536 689256 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
2537 689256 : pgstat_report_wait_end();
2538 689256 : if (written <= 0)
2539 : {
2540 : char xlogfname[MAXFNAMELEN];
2541 : int save_errno;
2542 :
2543 0 : if (errno == EINTR)
2544 0 : continue;
2545 :
2546 0 : save_errno = errno;
2547 0 : XLogFileName(xlogfname, ThisTimeLineID, openLogSegNo,
2548 : wal_segment_size);
2549 0 : errno = save_errno;
2550 0 : ereport(PANIC,
2551 : (errcode_for_file_access(),
2552 : errmsg("could not write to log file %s "
2553 : "at offset %u, length %zu: %m",
2554 : xlogfname, startoffset, nleft)));
2555 : }
2556 689256 : nleft -= written;
2557 689256 : from += written;
2558 689256 : startoffset += written;
2559 689256 : } while (nleft > 0);
2560 :
2561 689256 : npages = 0;
2562 :
2563 : /*
2564 : * If we just wrote the whole last page of a logfile segment,
2565 : * fsync the segment immediately. This avoids having to go back
2566 : * and re-open prior segments when an fsync request comes along
2567 : * later. Doing it here ensures that one and only one backend will
2568 : * perform this fsync.
2569 : *
2570 : * This is also the right place to notify the Archiver that the
2571 : * segment is ready to copy to archival storage, and to update the
2572 : * timer for archive_timeout, and to signal for a checkpoint if
2573 : * too many logfile segments have been used since the last
2574 : * checkpoint.
2575 : */
2576 689256 : if (finishing_seg)
2577 : {
2578 890 : issue_xlog_fsync(openLogFile, openLogSegNo);
2579 :
2580 : /* signal that we need to wakeup walsenders later */
2581 890 : WalSndWakeupRequest();
2582 :
2583 890 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2584 :
2585 890 : if (XLogArchivingActive())
2586 20 : XLogArchiveNotifySeg(openLogSegNo);
2587 :
2588 890 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
2589 890 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2590 :
2591 : /*
2592 : * Request a checkpoint if we've consumed too much xlog since
2593 : * the last one. For speed, we first check using the local
2594 : * copy of RedoRecPtr, which might be out of date; if it looks
2595 : * like a checkpoint is needed, forcibly update RedoRecPtr and
2596 : * recheck.
2597 : */
2598 890 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2599 : {
2600 60 : (void) GetRedoRecPtr();
2601 60 : if (XLogCheckpointNeeded(openLogSegNo))
2602 48 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2603 : }
2604 : }
2605 : }
2606 :
2607 1187012 : if (ispartialpage)
2608 : {
2609 : /* Only asked to write a partial page */
2610 262552 : LogwrtResult.Write = WriteRqst.Write;
2611 262552 : break;
2612 : }
2613 924460 : curridx = NextBufIdx(curridx);
2614 :
2615 : /* If flexible, break out of loop as soon as we wrote something */
2616 924460 : if (flexible && npages == 0)
2617 4978 : break;
2618 : }
2619 :
2620 : Assert(npages == 0);
2621 :
2622 : /*
2623 : * If asked to flush, do so
2624 : */
2625 687966 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2626 263874 : LogwrtResult.Flush < LogwrtResult.Write)
2627 :
2628 : {
2629 : /*
2630 : * Could get here without iterating above loop, in which case we might
2631 : * have no open file or the wrong one. However, we do not need to
2632 : * fsync more than one file.
2633 : */
2634 263822 : if (sync_method != SYNC_METHOD_OPEN &&
2635 263822 : sync_method != SYNC_METHOD_OPEN_DSYNC)
2636 : {
2637 263822 : if (openLogFile >= 0 &&
2638 263812 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2639 : wal_segment_size))
2640 0 : XLogFileClose();
2641 263822 : if (openLogFile < 0)
2642 : {
2643 10 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2644 : wal_segment_size);
2645 10 : openLogFile = XLogFileOpen(openLogSegNo);
2646 10 : ReserveExternalFD();
2647 : }
2648 :
2649 263822 : issue_xlog_fsync(openLogFile, openLogSegNo);
2650 : }
2651 :
2652 : /* signal that we need to wakeup walsenders later */
2653 263822 : WalSndWakeupRequest();
2654 :
2655 263822 : LogwrtResult.Flush = LogwrtResult.Write;
2656 : }
2657 :
2658 : /*
2659 : * Update shared-memory status
2660 : *
2661 : * We make sure that the shared 'request' values do not fall behind the
2662 : * 'result' values. This is not absolutely essential, but it saves some
2663 : * code in a couple of places.
2664 : */
2665 : {
2666 687966 : SpinLockAcquire(&XLogCtl->info_lck);
2667 687966 : XLogCtl->LogwrtResult = LogwrtResult;
2668 687966 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2669 250952 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2670 687966 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2671 264284 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2672 687966 : SpinLockRelease(&XLogCtl->info_lck);
2673 : }
2674 687966 : }
2675 :
2676 : /*
2677 : * Record the LSN for an asynchronous transaction commit/abort
2678 : * and nudge the WALWriter if there is work for it to do.
2679 : * (This should not be called for synchronous commits.)
2680 : */
2681 : void
2682 62948 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2683 : {
2684 62948 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2685 : bool sleeping;
2686 :
2687 62948 : SpinLockAcquire(&XLogCtl->info_lck);
2688 62948 : LogwrtResult = XLogCtl->LogwrtResult;
2689 62948 : sleeping = XLogCtl->WalWriterSleeping;
2690 62948 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2691 62668 : XLogCtl->asyncXactLSN = asyncXactLSN;
2692 62948 : SpinLockRelease(&XLogCtl->info_lck);
2693 :
2694 : /*
2695 : * If the WALWriter is sleeping, we should kick it to make it come out of
2696 : * low-power mode. Otherwise, determine whether there's a full page of
2697 : * WAL available to write.
2698 : */
2699 62948 : if (!sleeping)
2700 : {
2701 : /* back off to last completed page boundary */
2702 62916 : WriteRqstPtr -= WriteRqstPtr % XLOG_BLCKSZ;
2703 :
2704 : /* if we have already flushed that far, we're done */
2705 62916 : if (WriteRqstPtr <= LogwrtResult.Flush)
2706 13376 : return;
2707 : }
2708 :
2709 : /*
2710 : * Nudge the WALWriter: it has a full page of WAL to write, or we want it
2711 : * to come out of low-power mode so that this async commit will reach disk
2712 : * within the expected amount of time.
2713 : */
2714 49572 : if (ProcGlobal->walwriterLatch)
2715 11622 : SetLatch(ProcGlobal->walwriterLatch);
2716 : }
2717 :
2718 : /*
2719 : * Record the LSN up to which we can remove WAL because it's not required by
2720 : * any replication slot.
2721 : */
2722 : void
2723 2498 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2724 : {
2725 2498 : SpinLockAcquire(&XLogCtl->info_lck);
2726 2498 : XLogCtl->replicationSlotMinLSN = lsn;
2727 2498 : SpinLockRelease(&XLogCtl->info_lck);
2728 2498 : }
2729 :
2730 :
2731 : /*
2732 : * Return the oldest LSN we must retain to satisfy the needs of some
2733 : * replication slot.
2734 : */
2735 : static XLogRecPtr
2736 3344 : XLogGetReplicationSlotMinimumLSN(void)
2737 : {
2738 : XLogRecPtr retval;
2739 :
2740 3344 : SpinLockAcquire(&XLogCtl->info_lck);
2741 3344 : retval = XLogCtl->replicationSlotMinLSN;
2742 3344 : SpinLockRelease(&XLogCtl->info_lck);
2743 :
2744 3344 : return retval;
2745 : }
2746 :
2747 : /*
2748 : * Advance minRecoveryPoint in control file.
2749 : *
2750 : * If we crash during recovery, we must reach this point again before the
2751 : * database is consistent.
2752 : *
2753 : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2754 : * is only updated if it's not already greater than or equal to 'lsn'.
2755 : */
2756 : static void
2757 4188 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2758 : {
2759 : /* Quick check using our local copy of the variable */
2760 4188 : if (!updateMinRecoveryPoint || (!force && lsn <= minRecoveryPoint))
2761 3726 : return;
2762 :
2763 : /*
2764 : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2765 : * i.e., we're doing crash recovery. We never modify the control file's
2766 : * value in that case, so we can short-circuit future checks here too. The
2767 : * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2768 : * updated until crash recovery finishes. We only do this for the startup
2769 : * process as it should not update its own reference of minRecoveryPoint
2770 : * until it has finished crash recovery to make sure that all WAL
2771 : * available is replayed in this case. This also saves from extra locks
2772 : * taken on the control file from the startup process.
2773 : */
2774 462 : if (XLogRecPtrIsInvalid(minRecoveryPoint) && InRecovery)
2775 : {
2776 36 : updateMinRecoveryPoint = false;
2777 36 : return;
2778 : }
2779 :
2780 426 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2781 :
2782 : /* update local copy */
2783 426 : minRecoveryPoint = ControlFile->minRecoveryPoint;
2784 426 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2785 :
2786 426 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
2787 0 : updateMinRecoveryPoint = false;
2788 426 : else if (force || minRecoveryPoint < lsn)
2789 : {
2790 : XLogRecPtr newMinRecoveryPoint;
2791 : TimeLineID newMinRecoveryPointTLI;
2792 :
2793 : /*
2794 : * To avoid having to update the control file too often, we update it
2795 : * all the way to the last record being replayed, even though 'lsn'
2796 : * would suffice for correctness. This also allows the 'force' case
2797 : * to not need a valid 'lsn' value.
2798 : *
2799 : * Another important reason for doing it this way is that the passed
2800 : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2801 : * the caller got it from a corrupted heap page. Accepting such a
2802 : * value as the min recovery point would prevent us from coming up at
2803 : * all. Instead, we just log a warning and continue with recovery.
2804 : * (See also the comments about corrupt LSNs in XLogFlush.)
2805 : */
2806 362 : SpinLockAcquire(&XLogCtl->info_lck);
2807 362 : newMinRecoveryPoint = XLogCtl->replayEndRecPtr;
2808 362 : newMinRecoveryPointTLI = XLogCtl->replayEndTLI;
2809 362 : SpinLockRelease(&XLogCtl->info_lck);
2810 :
2811 362 : if (!force && newMinRecoveryPoint < lsn)
2812 0 : elog(WARNING,
2813 : "xlog min recovery request %X/%X is past current point %X/%X",
2814 : (uint32) (lsn >> 32), (uint32) lsn,
2815 : (uint32) (newMinRecoveryPoint >> 32),
2816 : (uint32) newMinRecoveryPoint);
2817 :
2818 : /* update control file */
2819 362 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2820 : {
2821 342 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
2822 342 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
2823 342 : UpdateControlFile();
2824 342 : minRecoveryPoint = newMinRecoveryPoint;
2825 342 : minRecoveryPointTLI = newMinRecoveryPointTLI;
2826 :
2827 342 : ereport(DEBUG2,
2828 : (errmsg("updated min recovery point to %X/%X on timeline %u",
2829 : (uint32) (minRecoveryPoint >> 32),
2830 : (uint32) minRecoveryPoint,
2831 : newMinRecoveryPointTLI)));
2832 : }
2833 : }
2834 426 : LWLockRelease(ControlFileLock);
2835 : }
2836 :
2837 : /*
2838 : * Ensure that all XLOG data through the given position is flushed to disk.
2839 : *
2840 : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2841 : * already held, and we try to avoid acquiring it if possible.
2842 : */
2843 : void
2844 788772 : XLogFlush(XLogRecPtr record)
2845 : {
2846 : XLogRecPtr WriteRqstPtr;
2847 : XLogwrtRqst WriteRqst;
2848 :
2849 : /*
2850 : * During REDO, we are reading not writing WAL. Therefore, instead of
2851 : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2852 : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2853 : * to act this way too, and because when it tries to write the
2854 : * end-of-recovery checkpoint, it should indeed flush.
2855 : */
2856 788772 : if (!XLogInsertAllowed())
2857 : {
2858 4094 : UpdateMinRecoveryPoint(record, false);
2859 524754 : return;
2860 : }
2861 :
2862 : /* Quick exit if already known flushed */
2863 784678 : if (record <= LogwrtResult.Flush)
2864 520660 : return;
2865 :
2866 : #ifdef WAL_DEBUG
2867 : if (XLOG_DEBUG)
2868 : elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
2869 : (uint32) (record >> 32), (uint32) record,
2870 : (uint32) (LogwrtResult.Write >> 32), (uint32) LogwrtResult.Write,
2871 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
2872 : #endif
2873 :
2874 264018 : START_CRIT_SECTION();
2875 :
2876 : /*
2877 : * Since fsync is usually a horribly expensive operation, we try to
2878 : * piggyback as much data as we can on each fsync: if we see any more data
2879 : * entered into the xlog buffer, we'll write and fsync that too, so that
2880 : * the final value of LogwrtResult.Flush is as large as possible. This
2881 : * gives us some chance of avoiding another fsync immediately after.
2882 : */
2883 :
2884 : /* initialize to given target; may increase below */
2885 264018 : WriteRqstPtr = record;
2886 :
2887 : /*
2888 : * Now wait until we get the write lock, or someone else does the flush
2889 : * for us.
2890 : */
2891 : for (;;)
2892 610 : {
2893 : XLogRecPtr insertpos;
2894 :
2895 : /* read LogwrtResult and update local state */
2896 264628 : SpinLockAcquire(&XLogCtl->info_lck);
2897 264628 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2898 2112 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2899 264628 : LogwrtResult = XLogCtl->LogwrtResult;
2900 264628 : SpinLockRelease(&XLogCtl->info_lck);
2901 :
2902 : /* done already? */
2903 264628 : if (record <= LogwrtResult.Flush)
2904 1384 : break;
2905 :
2906 : /*
2907 : * Before actually performing the write, wait for all in-flight
2908 : * insertions to the pages we're about to write to finish.
2909 : */
2910 263244 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2911 :
2912 : /*
2913 : * Try to get the write lock. If we can't get it immediately, wait
2914 : * until it's released, and recheck if we still need to do the flush
2915 : * or if the backend that held the lock did it for us already. This
2916 : * helps to maintain a good rate of group committing when the system
2917 : * is bottlenecked by the speed of fsyncing.
2918 : */
2919 263244 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2920 : {
2921 : /*
2922 : * The lock is now free, but we didn't acquire it yet. Before we
2923 : * do, loop back to check if someone else flushed the record for
2924 : * us already.
2925 : */
2926 610 : continue;
2927 : }
2928 :
2929 : /* Got the lock; recheck whether request is satisfied */
2930 262634 : LogwrtResult = XLogCtl->LogwrtResult;
2931 262634 : if (record <= LogwrtResult.Flush)
2932 : {
2933 138 : LWLockRelease(WALWriteLock);
2934 138 : break;
2935 : }
2936 :
2937 : /*
2938 : * Sleep before flush! By adding a delay here, we may give further
2939 : * backends the opportunity to join the backlog of group commit
2940 : * followers; this can significantly improve transaction throughput,
2941 : * at the risk of increasing transaction latency.
2942 : *
2943 : * We do not sleep if enableFsync is not turned on, nor if there are
2944 : * fewer than CommitSiblings other backends with active transactions.
2945 : */
2946 262496 : if (CommitDelay > 0 && enableFsync &&
2947 0 : MinimumActiveBackends(CommitSiblings))
2948 : {
2949 0 : pg_usleep(CommitDelay);
2950 :
2951 : /*
2952 : * Re-check how far we can now flush the WAL. It's generally not
2953 : * safe to call WaitXLogInsertionsToFinish while holding
2954 : * WALWriteLock, because an in-progress insertion might need to
2955 : * also grab WALWriteLock to make progress. But we know that all
2956 : * the insertions up to insertpos have already finished, because
2957 : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2958 : * We're only calling it again to allow insertpos to be moved
2959 : * further forward, not to actually wait for anyone.
2960 : */
2961 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2962 : }
2963 :
2964 : /* try to write/flush later additions to XLOG as well */
2965 262496 : WriteRqst.Write = insertpos;
2966 262496 : WriteRqst.Flush = insertpos;
2967 :
2968 262496 : XLogWrite(WriteRqst, false);
2969 :
2970 262496 : LWLockRelease(WALWriteLock);
2971 : /* done */
2972 262496 : break;
2973 : }
2974 :
2975 264018 : END_CRIT_SECTION();
2976 :
2977 : /* wake up walsenders now that we've released heavily contended locks */
2978 264018 : WalSndWakeupProcessRequests();
2979 :
2980 : /*
2981 : * If we still haven't flushed to the request point then we have a
2982 : * problem; most likely, the requested flush point is past end of XLOG.
2983 : * This has been seen to occur when a disk page has a corrupted LSN.
2984 : *
2985 : * Formerly we treated this as a PANIC condition, but that hurts the
2986 : * system's robustness rather than helping it: we do not want to take down
2987 : * the whole system due to corruption on one data page. In particular, if
2988 : * the bad page is encountered again during recovery then we would be
2989 : * unable to restart the database at all! (This scenario actually
2990 : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2991 : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2992 : * the only time we can reach here during recovery is while flushing the
2993 : * end-of-recovery checkpoint record, and we don't expect that to have a
2994 : * bad LSN.
2995 : *
2996 : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2997 : * since xact.c calls this routine inside a critical section. However,
2998 : * calls from bufmgr.c are not within critical sections and so we will not
2999 : * force a restart for a bad LSN on a data page.
3000 : */
3001 264018 : if (LogwrtResult.Flush < record)
3002 0 : elog(ERROR,
3003 : "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
3004 : (uint32) (record >> 32), (uint32) record,
3005 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
3006 : }
3007 :
3008 : /*
3009 : * Write & flush xlog, but without specifying exactly where to.
3010 : *
3011 : * We normally write only completed blocks; but if there is nothing to do on
3012 : * that basis, we check for unwritten async commits in the current incomplete
3013 : * block, and write through the latest one of those. Thus, if async commits
3014 : * are not being used, we will write complete blocks only.
3015 : *
3016 : * If, based on the above, there's anything to write we do so immediately. But
3017 : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
3018 : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
3019 : * more than wal_writer_flush_after unflushed blocks.
3020 : *
3021 : * We can guarantee that async commits reach disk after at most three
3022 : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
3023 : * to write "flexibly", meaning it can stop at the end of the buffer ring;
3024 : * this makes a difference only with very high load or long wal_writer_delay,
3025 : * but imposes one extra cycle for the worst case for async commits.)
3026 : *
3027 : * This routine is invoked periodically by the background walwriter process.
3028 : *
3029 : * Returns true if there was any work to do, even if we skipped flushing due
3030 : * to wal_writer_delay/wal_writer_flush_after.
3031 : */
3032 : bool
3033 55000 : XLogBackgroundFlush(void)
3034 : {
3035 : XLogwrtRqst WriteRqst;
3036 55000 : bool flexible = true;
3037 : static TimestampTz lastflush;
3038 : TimestampTz now;
3039 : int flushbytes;
3040 :
3041 : /* XLOG doesn't need flushing during recovery */
3042 55000 : if (RecoveryInProgress())
3043 0 : return false;
3044 :
3045 : /* read LogwrtResult and update local state */
3046 55000 : SpinLockAcquire(&XLogCtl->info_lck);
3047 55000 : LogwrtResult = XLogCtl->LogwrtResult;
3048 55000 : WriteRqst = XLogCtl->LogwrtRqst;
3049 55000 : SpinLockRelease(&XLogCtl->info_lck);
3050 :
3051 : /* back off to last completed page boundary */
3052 55000 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3053 :
3054 : /* if we have already flushed that far, consider async commit records */
3055 55000 : if (WriteRqst.Write <= LogwrtResult.Flush)
3056 : {
3057 46096 : SpinLockAcquire(&XLogCtl->info_lck);
3058 46096 : WriteRqst.Write = XLogCtl->asyncXactLSN;
3059 46096 : SpinLockRelease(&XLogCtl->info_lck);
3060 46096 : flexible = false; /* ensure it all gets written */
3061 : }
3062 :
3063 : /*
3064 : * If already known flushed, we're done. Just need to check if we are
3065 : * holding an open file handle to a logfile that's no longer in use,
3066 : * preventing the file from being deleted.
3067 : */
3068 55000 : if (WriteRqst.Write <= LogwrtResult.Flush)
3069 : {
3070 45650 : if (openLogFile >= 0)
3071 : {
3072 4706 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3073 : wal_segment_size))
3074 : {
3075 98 : XLogFileClose();
3076 : }
3077 : }
3078 45650 : return false;
3079 : }
3080 :
3081 : /*
3082 : * Determine how far to flush WAL, based on the wal_writer_delay and
3083 : * wal_writer_flush_after GUCs.
3084 : */
3085 9350 : now = GetCurrentTimestamp();
3086 9350 : flushbytes =
3087 9350 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3088 :
3089 9350 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3090 : {
3091 : /* first call, or block based limits disabled */
3092 164 : WriteRqst.Flush = WriteRqst.Write;
3093 164 : lastflush = now;
3094 : }
3095 9186 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3096 : {
3097 : /*
3098 : * Flush the writes at least every WalWriterDelay ms. This is
3099 : * important to bound the amount of time it takes for an asynchronous
3100 : * commit to hit disk.
3101 : */
3102 1526 : WriteRqst.Flush = WriteRqst.Write;
3103 1526 : lastflush = now;
3104 : }
3105 7660 : else if (flushbytes >= WalWriterFlushAfter)
3106 : {
3107 : /* exceeded wal_writer_flush_after blocks, flush */
3108 0 : WriteRqst.Flush = WriteRqst.Write;
3109 0 : lastflush = now;
3110 : }
3111 : else
3112 : {
3113 : /* no flushing, this time round */
3114 7660 : WriteRqst.Flush = 0;
3115 : }
3116 :
3117 : #ifdef WAL_DEBUG
3118 : if (XLOG_DEBUG)
3119 : elog(LOG, "xlog bg flush request write %X/%X; flush: %X/%X, current is write %X/%X; flush %X/%X",
3120 : (uint32) (WriteRqst.Write >> 32), (uint32) WriteRqst.Write,
3121 : (uint32) (WriteRqst.Flush >> 32), (uint32) WriteRqst.Flush,
3122 : (uint32) (LogwrtResult.Write >> 32), (uint32) LogwrtResult.Write,
3123 : (uint32) (LogwrtResult.Flush >> 32), (uint32) LogwrtResult.Flush);
3124 : #endif
3125 :
3126 9350 : START_CRIT_SECTION();
3127 :
3128 : /* now wait for any in-progress insertions to finish and get write lock */
3129 9350 : WaitXLogInsertionsToFinish(WriteRqst.Write);
3130 9350 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3131 9350 : LogwrtResult = XLogCtl->LogwrtResult;
3132 9350 : if (WriteRqst.Write > LogwrtResult.Write ||
3133 3930 : WriteRqst.Flush > LogwrtResult.Flush)
3134 : {
3135 5476 : XLogWrite(WriteRqst, flexible);
3136 : }
3137 9350 : LWLockRelease(WALWriteLock);
3138 :
3139 9350 : END_CRIT_SECTION();
3140 :
3141 : /* wake up walsenders now that we've released heavily contended locks */
3142 9350 : WalSndWakeupProcessRequests();
3143 :
3144 : /*
3145 : * Great, done. To take some work off the critical path, try to initialize
3146 : * as many of the no-longer-needed WAL buffers for future use as we can.
3147 : */
3148 9350 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, true);
3149 :
3150 : /*
3151 : * If we determined that we need to write data, but somebody else
3152 : * wrote/flushed already, it should be considered as being active, to
3153 : * avoid hibernating too early.
3154 : */
3155 9350 : return true;
3156 : }
3157 :
3158 : /*
3159 : * Test whether XLOG data has been flushed up to (at least) the given position.
3160 : *
3161 : * Returns true if a flush is still needed. (It may be that someone else
3162 : * is already in process of flushing that far, however.)
3163 : */
3164 : bool
3165 16995374 : XLogNeedsFlush(XLogRecPtr record)
3166 : {
3167 : /*
3168 : * During recovery, we don't flush WAL but update minRecoveryPoint
3169 : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3170 : * would need to be updated.
3171 : */
3172 16995374 : if (RecoveryInProgress())
3173 : {
3174 : /*
3175 : * An invalid minRecoveryPoint means that we need to recover all the
3176 : * WAL, i.e., we're doing crash recovery. We never modify the control
3177 : * file's value in that case, so we can short-circuit future checks
3178 : * here too. This triggers a quick exit path for the startup process,
3179 : * which cannot update its local copy of minRecoveryPoint as long as
3180 : * it has not replayed all WAL available when doing crash recovery.
3181 : */
3182 132956 : if (XLogRecPtrIsInvalid(minRecoveryPoint) && InRecovery)
3183 0 : updateMinRecoveryPoint = false;
3184 :
3185 : /* Quick exit if already known to be updated or cannot be updated */
3186 132956 : if (record <= minRecoveryPoint || !updateMinRecoveryPoint)
3187 103700 : return false;
3188 :
3189 : /*
3190 : * Update local copy of minRecoveryPoint. But if the lock is busy,
3191 : * just return a conservative guess.
3192 : */
3193 29256 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3194 0 : return true;
3195 29256 : minRecoveryPoint = ControlFile->minRecoveryPoint;
3196 29256 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
3197 29256 : LWLockRelease(ControlFileLock);
3198 :
3199 : /*
3200 : * Check minRecoveryPoint for any other process than the startup
3201 : * process doing crash recovery, which should not update the control
3202 : * file value if crash recovery is still running.
3203 : */
3204 29256 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
3205 0 : updateMinRecoveryPoint = false;
3206 :
3207 : /* check again */
3208 29256 : if (record <= minRecoveryPoint || !updateMinRecoveryPoint)
3209 28 : return false;
3210 : else
3211 29228 : return true;
3212 : }
3213 :
3214 : /* Quick exit if already known flushed */
3215 16862418 : if (record <= LogwrtResult.Flush)
3216 16783410 : return false;
3217 :
3218 : /* read LogwrtResult and update local state */
3219 79008 : SpinLockAcquire(&XLogCtl->info_lck);
3220 79008 : LogwrtResult = XLogCtl->LogwrtResult;
3221 79008 : SpinLockRelease(&XLogCtl->info_lck);
3222 :
3223 : /* check again */
3224 79008 : if (record <= LogwrtResult.Flush)
3225 1056 : return false;
3226 :
3227 77952 : return true;
3228 : }
3229 :
3230 : /*
3231 : * Create a new XLOG file segment, or open a pre-existing one.
3232 : *
3233 : * logsegno: identify segment to be created/opened.
3234 : *
3235 : * *use_existent: if true, OK to use a pre-existing file (else, any
3236 : * pre-existing file will be deleted). On return, true if a pre-existing
3237 : * file was used.
3238 : *
3239 : * use_lock: if true, acquire ControlFileLock while moving file into
3240 : * place. This should be true except during bootstrap log creation. The
3241 : * caller must *not* hold the lock at call.
3242 : *
3243 : * Returns FD of opened file.
3244 : *
3245 : * Note: errors here are ERROR not PANIC because we might or might not be
3246 : * inside a critical section (eg, during checkpoint there is no reason to
3247 : * take down the system on failure). They will promote to PANIC if we are
3248 : * in a critical section.
3249 : */
3250 : int
3251 8782 : XLogFileInit(XLogSegNo logsegno, bool *use_existent, bool use_lock)
3252 : {
3253 : char path[MAXPGPATH];
3254 : char tmppath[MAXPGPATH];
3255 : PGAlignedXLogBlock zbuffer;
3256 : XLogSegNo installed_segno;
3257 : XLogSegNo max_segno;
3258 : int fd;
3259 : int nbytes;
3260 : int save_errno;
3261 :
3262 8782 : XLogFilePath(path, ThisTimeLineID, logsegno, wal_segment_size);
3263 :
3264 : /*
3265 : * Try to use existent file (checkpoint maker may have created it already)
3266 : */
3267 8782 : if (*use_existent)
3268 : {
3269 8424 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method));
3270 8424 : if (fd < 0)
3271 : {
3272 418 : if (errno != ENOENT)
3273 0 : ereport(ERROR,
3274 : (errcode_for_file_access(),
3275 : errmsg("could not open file \"%s\": %m", path)));
3276 : }
3277 : else
3278 8006 : return fd;
3279 : }
3280 :
3281 : /*
3282 : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3283 : * another process is doing the same thing. If so, we will end up
3284 : * pre-creating an extra log segment. That seems OK, and better than
3285 : * holding the lock throughout this lengthy process.
3286 : */
3287 776 : elog(DEBUG2, "creating and filling new WAL file");
3288 :
3289 776 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3290 :
3291 776 : unlink(tmppath);
3292 :
3293 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3294 776 : fd = BasicOpenFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3295 776 : if (fd < 0)
3296 0 : ereport(ERROR,
3297 : (errcode_for_file_access(),
3298 : errmsg("could not create file \"%s\": %m", tmppath)));
3299 :
3300 776 : memset(zbuffer.data, 0, XLOG_BLCKSZ);
3301 :
3302 776 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3303 776 : save_errno = 0;
3304 776 : if (wal_init_zero)
3305 : {
3306 : /*
3307 : * Zero-fill the file. With this setting, we do this the hard way to
3308 : * ensure that all the file space has really been allocated. On
3309 : * platforms that allow "holes" in files, just seeking to the end
3310 : * doesn't allocate intermediate space. This way, we know that we
3311 : * have all the space and (after the fsync below) that all the
3312 : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3313 : * O_DSYNC will be sufficient to sync future writes to the log file.
3314 : */
3315 1152264 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += XLOG_BLCKSZ)
3316 : {
3317 1151488 : errno = 0;
3318 1151488 : if (write(fd, zbuffer.data, XLOG_BLCKSZ) != XLOG_BLCKSZ)
3319 : {
3320 : /* if write didn't set errno, assume no disk space */
3321 0 : save_errno = errno ? errno : ENOSPC;
3322 0 : break;
3323 : }
3324 : }
3325 : }
3326 : else
3327 : {
3328 : /*
3329 : * Otherwise, seeking to the end and writing a solitary byte is
3330 : * enough.
3331 : */
3332 0 : errno = 0;
3333 0 : if (pg_pwrite(fd, zbuffer.data, 1, wal_segment_size - 1) != 1)
3334 : {
3335 : /* if write didn't set errno, assume no disk space */
3336 0 : save_errno = errno ? errno : ENOSPC;
3337 : }
3338 : }
3339 776 : pgstat_report_wait_end();
3340 :
3341 776 : if (save_errno)
3342 : {
3343 : /*
3344 : * If we fail to make the file, delete it to release disk space
3345 : */
3346 0 : unlink(tmppath);
3347 :
3348 0 : close(fd);
3349 :
3350 0 : errno = save_errno;
3351 :
3352 0 : ereport(ERROR,
3353 : (errcode_for_file_access(),
3354 : errmsg("could not write to file \"%s\": %m", tmppath)));
3355 : }
3356 :
3357 776 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
3358 776 : if (pg_fsync(fd) != 0)
3359 : {
3360 0 : int save_errno = errno;
3361 :
3362 0 : close(fd);
3363 0 : errno = save_errno;
3364 0 : ereport(ERROR,
3365 : (errcode_for_file_access(),
3366 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3367 : }
3368 776 : pgstat_report_wait_end();
3369 :
3370 776 : if (close(fd) != 0)
3371 0 : ereport(ERROR,
3372 : (errcode_for_file_access(),
3373 : errmsg("could not close file \"%s\": %m", tmppath)));
3374 :
3375 : /*
3376 : * Now move the segment into place with its final name.
3377 : *
3378 : * If caller didn't want to use a pre-existing file, get rid of any
3379 : * pre-existing file. Otherwise, cope with possibility that someone else
3380 : * has created the file while we were filling ours: if so, use ours to
3381 : * pre-create a future log segment.
3382 : */
3383 776 : installed_segno = logsegno;
3384 :
3385 : /*
3386 : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3387 : * that was a constant, but that was always a bit dubious: normally, at a
3388 : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3389 : * here, it was the offset from the insert location. We can't do the
3390 : * normal XLOGfileslop calculation here because we don't have access to
3391 : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3392 : * CheckPointSegments.
3393 : */
3394 776 : max_segno = logsegno + CheckPointSegments;
3395 776 : if (!InstallXLogFileSegment(&installed_segno, tmppath,
3396 776 : *use_existent, max_segno,
3397 : use_lock))
3398 : {
3399 : /*
3400 : * No need for any more future segments, or InstallXLogFileSegment()
3401 : * failed to rename the file into place. If the rename failed, opening
3402 : * the file below will fail.
3403 : */
3404 0 : unlink(tmppath);
3405 : }
3406 :
3407 : /* Set flag to tell caller there was no existent file */
3408 776 : *use_existent = false;
3409 :
3410 : /* Now open original target segment (might not be file I just made) */
3411 776 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method));
3412 776 : if (fd < 0)
3413 0 : ereport(ERROR,
3414 : (errcode_for_file_access(),
3415 : errmsg("could not open file \"%s\": %m", path)));
3416 :
3417 776 : elog(DEBUG2, "done creating and filling new WAL file");
3418 :
3419 776 : return fd;
3420 : }
3421 :
3422 : /*
3423 : * Create a new XLOG file segment by copying a pre-existing one.
3424 : *
3425 : * destsegno: identify segment to be created.
3426 : *
3427 : * srcTLI, srcsegno: identify segment to be copied (could be from
3428 : * a different timeline)
3429 : *
3430 : * upto: how much of the source file to copy (the rest is filled with
3431 : * zeros)
3432 : *
3433 : * Currently this is only used during recovery, and so there are no locking
3434 : * considerations. But we should be just as tense as XLogFileInit to avoid
3435 : * emplacing a bogus file.
3436 : */
3437 : static void
3438 36 : XLogFileCopy(XLogSegNo destsegno, TimeLineID srcTLI, XLogSegNo srcsegno,
3439 : int upto)
3440 : {
3441 : char path[MAXPGPATH];
3442 : char tmppath[MAXPGPATH];
3443 : PGAlignedXLogBlock buffer;
3444 : int srcfd;
3445 : int fd;
3446 : int nbytes;
3447 :
3448 : /*
3449 : * Open the source file
3450 : */
3451 36 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
3452 36 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
3453 36 : if (srcfd < 0)
3454 0 : ereport(ERROR,
3455 : (errcode_for_file_access(),
3456 : errmsg("could not open file \"%s\": %m", path)));
3457 :
3458 : /*
3459 : * Copy into a temp file name.
3460 : */
3461 36 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3462 :
3463 36 : unlink(tmppath);
3464 :
3465 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3466 36 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3467 36 : if (fd < 0)
3468 0 : ereport(ERROR,
3469 : (errcode_for_file_access(),
3470 : errmsg("could not create file \"%s\": %m", tmppath)));
3471 :
3472 : /*
3473 : * Do the data copying.
3474 : */
3475 73764 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3476 : {
3477 : int nread;
3478 :
3479 73728 : nread = upto - nbytes;
3480 :
3481 : /*
3482 : * The part that is not read from the source file is filled with
3483 : * zeros.
3484 : */
3485 73728 : if (nread < sizeof(buffer))
3486 36 : memset(buffer.data, 0, sizeof(buffer));
3487 :
3488 73728 : if (nread > 0)
3489 : {
3490 : int r;
3491 :
3492 1740 : if (nread > sizeof(buffer))
3493 1704 : nread = sizeof(buffer);
3494 1740 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
3495 1740 : r = read(srcfd, buffer.data, nread);
3496 1740 : if (r != nread)
3497 : {
3498 0 : if (r < 0)
3499 0 : ereport(ERROR,
3500 : (errcode_for_file_access(),
3501 : errmsg("could not read file \"%s\": %m",
3502 : path)));
3503 : else
3504 0 : ereport(ERROR,
3505 : (errcode(ERRCODE_DATA_CORRUPTED),
3506 : errmsg("could not read file \"%s\": read %d of %zu",
3507 : path, r, (Size) nread)));
3508 : }
3509 1740 : pgstat_report_wait_end();
3510 : }
3511 73728 : errno = 0;
3512 73728 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
3513 73728 : if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3514 : {
3515 0 : int save_errno = errno;
3516 :
3517 : /*
3518 : * If we fail to make the file, delete it to release disk space
3519 : */
3520 0 : unlink(tmppath);
3521 : /* if write didn't set errno, assume problem is no disk space */
3522 0 : errno = save_errno ? save_errno : ENOSPC;
3523 :
3524 0 : ereport(ERROR,
3525 : (errcode_for_file_access(),
3526 : errmsg("could not write to file \"%s\": %m", tmppath)));
3527 : }
3528 73728 : pgstat_report_wait_end();
3529 : }
3530 :
3531 36 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
3532 36 : if (pg_fsync(fd) != 0)
3533 0 : ereport(data_sync_elevel(ERROR),
3534 : (errcode_for_file_access(),
3535 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3536 36 : pgstat_report_wait_end();
3537 :
3538 36 : if (CloseTransientFile(fd) != 0)
3539 0 : ereport(ERROR,
3540 : (errcode_for_file_access(),
3541 : errmsg("could not close file \"%s\": %m", tmppath)));
3542 :
3543 36 : if (CloseTransientFile(srcfd) != 0)
3544 0 : ereport(ERROR,
3545 : (errcode_for_file_access(),
3546 : errmsg("could not close file \"%s\": %m", path)));
3547 :
3548 : /*
3549 : * Now move the segment into place with its final name.
3550 : */
3551 36 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, false))
3552 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
3553 36 : }
3554 :
3555 : /*
3556 : * Install a new XLOG segment file as a current or future log segment.
3557 : *
3558 : * This is used both to install a newly-created segment (which has a temp
3559 : * filename while it's being created) and to recycle an old segment.
3560 : *
3561 : * *segno: identify segment to install as (or first possible target).
3562 : * When find_free is true, this is modified on return to indicate the
3563 : * actual installation location or last segment searched.
3564 : *
3565 : * tmppath: initial name of file to install. It will be renamed into place.
3566 : *
3567 : * find_free: if true, install the new segment at the first empty segno
3568 : * number at or after the passed numbers. If false, install the new segment
3569 : * exactly where specified, deleting any existing segment file there.
3570 : *
3571 : * max_segno: maximum segment number to install the new file as. Fail if no
3572 : * free slot is found between *segno and max_segno. (Ignored when find_free
3573 : * is false.)
3574 : *
3575 : * use_lock: if true, acquire ControlFileLock while moving file into
3576 : * place. This should be true except during bootstrap log creation. The
3577 : * caller must *not* hold the lock at call.
3578 : *
3579 : * Returns true if the file was installed successfully. false indicates that
3580 : * max_segno limit was exceeded, or an error occurred while renaming the
3581 : * file into place.
3582 : */
3583 : static bool
3584 1544 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3585 : bool find_free, XLogSegNo max_segno,
3586 : bool use_lock)
3587 : {
3588 : char path[MAXPGPATH];
3589 : struct stat stat_buf;
3590 :
3591 1544 : XLogFilePath(path, ThisTimeLineID, *segno, wal_segment_size);
3592 :
3593 : /*
3594 : * We want to be sure that only one process does this at a time.
3595 : */
3596 1544 : if (use_lock)
3597 1150 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3598 :
3599 1544 : if (!find_free)
3600 : {
3601 : /* Force installation: get rid of any pre-existing segment file */
3602 394 : durable_unlink(path, DEBUG1);
3603 : }
3604 : else
3605 : {
3606 : /* Find a free slot to put it in */
3607 13480 : while (stat(path, &stat_buf) == 0)
3608 : {
3609 12350 : if ((*segno) >= max_segno)
3610 : {
3611 : /* Failed to find a free slot within specified range */
3612 20 : if (use_lock)
3613 20 : LWLockRelease(ControlFileLock);
3614 20 : return false;
3615 : }
3616 12330 : (*segno)++;
3617 12330 : XLogFilePath(path, ThisTimeLineID, *segno, wal_segment_size);
3618 : }
3619 : }
3620 :
3621 : /*
3622 : * Perform the rename using link if available, paranoidly trying to avoid
3623 : * overwriting an existing file (there shouldn't be one).
3624 : */
3625 1524 : if (durable_rename_excl(tmppath, path, LOG) != 0)
3626 : {
3627 0 : if (use_lock)
3628 0 : LWLockRelease(ControlFileLock);
3629 : /* durable_rename_excl already emitted log message */
3630 0 : return false;
3631 : }
3632 :
3633 1524 : if (use_lock)
3634 1130 : LWLockRelease(ControlFileLock);
3635 :
3636 1524 : return true;
3637 : }
3638 :
3639 : /*
3640 : * Open a pre-existing logfile segment for writing.
3641 : */
3642 : int
3643 10 : XLogFileOpen(XLogSegNo segno)
3644 : {
3645 : char path[MAXPGPATH];
3646 : int fd;
3647 :
3648 10 : XLogFilePath(path, ThisTimeLineID, segno, wal_segment_size);
3649 :
3650 10 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | get_sync_bit(sync_method));
3651 10 : if (fd < 0)
3652 0 : ereport(PANIC,
3653 : (errcode_for_file_access(),
3654 : errmsg("could not open file \"%s\": %m", path)));
3655 :
3656 10 : return fd;
3657 : }
3658 :
3659 : /*
3660 : * Open a logfile segment for reading (during recovery).
3661 : *
3662 : * If source == XLOG_FROM_ARCHIVE, the segment is retrieved from archive.
3663 : * Otherwise, it's assumed to be already available in pg_wal.
3664 : */
3665 : static int
3666 2308 : XLogFileRead(XLogSegNo segno, int emode, TimeLineID tli,
3667 : XLogSource source, bool notfoundOk)
3668 : {
3669 : char xlogfname[MAXFNAMELEN];
3670 : char activitymsg[MAXFNAMELEN + 16];
3671 : char path[MAXPGPATH];
3672 : int fd;
3673 :
3674 2308 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
3675 :
3676 2308 : switch (source)
3677 : {
3678 356 : case XLOG_FROM_ARCHIVE:
3679 : /* Report recovery progress in PS display */
3680 356 : snprintf(activitymsg, sizeof(activitymsg), "waiting for %s",
3681 : xlogfname);
3682 356 : set_ps_display(activitymsg);
3683 :
3684 356 : restoredFromArchive = RestoreArchivedFile(path, xlogfname,
3685 : "RECOVERYXLOG",
3686 : wal_segment_size,
3687 : InRedo);
3688 356 : if (!restoredFromArchive)
3689 338 : return -1;
3690 18 : break;
3691 :
3692 1952 : case XLOG_FROM_PG_WAL:
3693 : case XLOG_FROM_STREAM:
3694 1952 : XLogFilePath(path, tli, segno, wal_segment_size);
3695 1952 : restoredFromArchive = false;
3696 1952 : break;
3697 :
3698 0 : default:
3699 0 : elog(ERROR, "invalid XLogFileRead source %d", source);
3700 : }
3701 :
3702 : /*
3703 : * If the segment was fetched from archival storage, replace the existing
3704 : * xlog segment (if any) with the archival version.
3705 : */
3706 1970 : if (source == XLOG_FROM_ARCHIVE)
3707 : {
3708 18 : KeepFileRestoredFromArchive(path, xlogfname);
3709 :
3710 : /*
3711 : * Set path to point at the new file in pg_wal.
3712 : */
3713 18 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlogfname);
3714 : }
3715 :
3716 1970 : fd = BasicOpenFile(path, O_RDONLY | PG_BINARY);
3717 1970 : if (fd >= 0)
3718 : {
3719 : /* Success! */
3720 1856 : curFileTLI = tli;
3721 :
3722 : /* Report recovery progress in PS display */
3723 1856 : snprintf(activitymsg, sizeof(activitymsg), "recovering %s",
3724 : xlogfname);
3725 1856 : set_ps_display(activitymsg);
3726 :
3727 : /* Track source of data in assorted state variables */
3728 1856 : readSource = source;
3729 1856 : XLogReceiptSource = source;
3730 : /* In FROM_STREAM case, caller tracks receipt time, not me */
3731 1856 : if (source != XLOG_FROM_STREAM)
3732 1708 : XLogReceiptTime = GetCurrentTimestamp();
3733 :
3734 1856 : return fd;
3735 : }
3736 114 : if (errno != ENOENT || !notfoundOk) /* unexpected failure? */
3737 0 : ereport(PANIC,
3738 : (errcode_for_file_access(),
3739 : errmsg("could not open file \"%s\": %m", path)));
3740 114 : return -1;
3741 : }
3742 :
3743 : /*
3744 : * Open a logfile segment for reading (during recovery).
3745 : *
3746 : * This version searches for the segment with any TLI listed in expectedTLEs.
3747 : */
3748 : static int
3749 1810 : XLogFileReadAnyTLI(XLogSegNo segno, int emode, XLogSource source)
3750 : {
3751 : char path[MAXPGPATH];
3752 : ListCell *cell;
3753 : int fd;
3754 : List *tles;
3755 :
3756 : /*
3757 : * Loop looking for a suitable timeline ID: we might need to read any of
3758 : * the timelines listed in expectedTLEs.
3759 : *
3760 : * We expect curFileTLI on entry to be the TLI of the preceding file in
3761 : * sequence, or 0 if there was no predecessor. We do not allow curFileTLI
3762 : * to go backwards; this prevents us from picking up the wrong file when a
3763 : * parent timeline extends to higher segment numbers than the child we
3764 : * want to read.
3765 : *
3766 : * If we haven't read the timeline history file yet, read it now, so that
3767 : * we know which TLIs to scan. We don't save the list in expectedTLEs,
3768 : * however, unless we actually find a valid segment. That way if there is
3769 : * neither a timeline history file nor a WAL segment in the archive, and
3770 : * streaming replication is set up, we'll read the timeline history file
3771 : * streamed from the master when we start streaming, instead of recovering
3772 : * with a dummy history generated here.
3773 : */
3774 1810 : if (expectedTLEs)
3775 420 : tles = expectedTLEs;
3776 : else
3777 1390 : tles = readTimeLineHistory(recoveryTargetTLI);
3778 :
3779 1932 : foreach(cell, tles)
3780 : {
3781 1830 : TimeLineHistoryEntry *hent = (TimeLineHistoryEntry *) lfirst(cell);
3782 1830 : TimeLineID tli = hent->tli;
3783 :
3784 1830 : if (tli < curFileTLI)
3785 0 : break; /* don't bother looking at too-old TLIs */
3786 :
3787 : /*
3788 : * Skip scanning the timeline ID that the logfile segment to read
3789 : * doesn't belong to
3790 : */
3791 1830 : if (hent->begin != InvalidXLogRecPtr)
3792 : {
3793 70 : XLogSegNo beginseg = 0;
3794 :
3795 70 : XLByteToSeg(hent->begin, beginseg, wal_segment_size);
3796 :
3797 : /*
3798 : * The logfile segment that doesn't belong to the timeline is
3799 : * older or newer than the segment that the timeline started or
3800 : * ended at, respectively. It's sufficient to check only the
3801 : * starting segment of the timeline here. Since the timelines are
3802 : * scanned in descending order in this loop, any segments newer
3803 : * than the ending segment should belong to newer timeline and
3804 : * have already been read before. So it's not necessary to check
3805 : * the ending segment of the timeline here.
3806 : */
3807 70 : if (segno < beginseg)
3808 8 : continue;
3809 : }
3810 :
3811 1822 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_ARCHIVE)
3812 : {
3813 356 : fd = XLogFileRead(segno, emode, tli,
3814 : XLOG_FROM_ARCHIVE, true);
3815 356 : if (fd != -1)
3816 : {
3817 18 : elog(DEBUG1, "got WAL segment from archive");
3818 18 : if (!expectedTLEs)
3819 8 : expectedTLEs = tles;
3820 1708 : return fd;
3821 : }
3822 : }
3823 :
3824 1804 : if (source == XLOG_FROM_ANY || source == XLOG_FROM_PG_WAL)
3825 : {
3826 1804 : fd = XLogFileRead(segno, emode, tli,
3827 : XLOG_FROM_PG_WAL, true);
3828 1804 : if (fd != -1)
3829 : {
3830 1690 : if (!expectedTLEs)
3831 1382 : expectedTLEs = tles;
3832 1690 : return fd;
3833 : }
3834 : }
3835 : }
3836 :
3837 : /* Couldn't find it. For simplicity, complain about front timeline */
3838 102 : XLogFilePath(path, recoveryTargetTLI, segno, wal_segment_size);
3839 102 : errno = ENOENT;
3840 102 : ereport(emode,
3841 : (errcode_for_file_access(),
3842 : errmsg("could not open file \"%s\": %m", path)));
3843 102 : return -1;
3844 : }
3845 :
3846 : /*
3847 : * Close the current logfile segment for writing.
3848 : */
3849 : static void
3850 2042 : XLogFileClose(void)
3851 : {
3852 : Assert(openLogFile >= 0);
3853 :
3854 : /*
3855 : * WAL segment files will not be re-read in normal operation, so we advise
3856 : * the OS to release any cached pages. But do not do so if WAL archiving
3857 : * or streaming is active, because archiver and walsender process could
3858 : * use the cache to read the WAL segment.
3859 : */
3860 : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
3861 2042 : if (!XLogIsNeeded())
3862 40 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3863 : #endif
3864 :
3865 2042 : if (close(openLogFile) != 0)
3866 : {
3867 : char xlogfname[MAXFNAMELEN];
3868 0 : int save_errno = errno;
3869 :
3870 0 : XLogFileName(xlogfname, ThisTimeLineID, openLogSegNo, wal_segment_size);
3871 0 : errno = save_errno;
3872 0 : ereport(PANIC,
3873 : (errcode_for_file_access(),
3874 : errmsg("could not close file \"%s\": %m", xlogfname)));
3875 : }
3876 :
3877 2042 : openLogFile = -1;
3878 2042 : ReleaseExternalFD();
3879 2042 : }
3880 :
3881 : /*
3882 : * Preallocate log files beyond the specified log endpoint.
3883 : *
3884 : * XXX this is currently extremely conservative, since it forces only one
3885 : * future log segment to exist, and even that only if we are 75% done with
3886 : * the current one. This is only appropriate for very low-WAL-volume systems.
3887 : * High-volume systems will be OK once they've built up a sufficient set of
3888 : * recycled log segments, but the startup transient is likely to include
3889 : * a lot of segment creations by foreground processes, which is not so good.
3890 : */
3891 : static void
3892 3320 : PreallocXlogFiles(XLogRecPtr endptr)
3893 : {
3894 : XLogSegNo _logSegNo;
3895 : int lf;
3896 : bool use_existent;
3897 : uint64 offset;
3898 :
3899 3320 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3900 3320 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3901 3320 : if (offset >= (uint32) (0.75 * wal_segment_size))
3902 : {
3903 34 : _logSegNo++;
3904 34 : use_existent = true;
3905 34 : lf = XLogFileInit(_logSegNo, &use_existent, true);
3906 34 : close(lf);
3907 34 : if (!use_existent)
3908 4 : CheckpointStats.ckpt_segs_added++;
3909 : }
3910 3320 : }
3911 :
3912 : /*
3913 : * Throws an error if the given log segment has already been removed or
3914 : * recycled. The caller should only pass a segment that it knows to have
3915 : * existed while the server has been running, as this function always
3916 : * succeeds if no WAL segments have been removed since startup.
3917 : * 'tli' is only used in the error message.
3918 : *
3919 : * Note: this function guarantees to keep errno unchanged on return.
3920 : * This supports callers that use this to possibly deliver a better
3921 : * error message about a missing file, while still being able to throw
3922 : * a normal file-access error afterwards, if this does return.
3923 : */
3924 : void
3925 45254 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3926 : {
3927 45254 : int save_errno = errno;
3928 : XLogSegNo lastRemovedSegNo;
3929 :
3930 45254 : SpinLockAcquire(&XLogCtl->info_lck);
3931 45254 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3932 45254 : SpinLockRelease(&XLogCtl->info_lck);
3933 :
3934 45254 : if (segno <= lastRemovedSegNo)
3935 : {
3936 : char filename[MAXFNAMELEN];
3937 :
3938 0 : XLogFileName(filename, tli, segno, wal_segment_size);
3939 0 : errno = save_errno;
3940 0 : ereport(ERROR,
3941 : (errcode_for_file_access(),
3942 : errmsg("requested WAL segment %s has already been removed",
3943 : filename)));
3944 : }
3945 45254 : errno = save_errno;
3946 45254 : }
3947 :
3948 : /*
3949 : * Return the last WAL segment removed, or 0 if no segment has been removed
3950 : * since startup.
3951 : *
3952 : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3953 : * with that.
3954 : */
3955 : XLogSegNo
3956 592 : XLogGetLastRemovedSegno(void)
3957 : {
3958 : XLogSegNo lastRemovedSegNo;
3959 :
3960 592 : SpinLockAcquire(&XLogCtl->info_lck);
3961 592 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3962 592 : SpinLockRelease(&XLogCtl->info_lck);
3963 :
3964 592 : return lastRemovedSegNo;
3965 : }
3966 :
3967 :
3968 : /*
3969 : * Update the last removed segno pointer in shared memory, to reflect that the
3970 : * given XLOG file has been removed.
3971 : */
3972 : static void
3973 718 : UpdateLastRemovedPtr(char *filename)
3974 : {
3975 : uint32 tli;
3976 : XLogSegNo segno;
3977 :
3978 718 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3979 :
3980 718 : SpinLockAcquire(&XLogCtl->info_lck);
3981 718 : if (segno > XLogCtl->lastRemovedSegNo)
3982 194 : XLogCtl->lastRemovedSegNo = segno;
3983 718 : SpinLockRelease(&XLogCtl->info_lck);
3984 718 : }
3985 :
3986 : /*
3987 : * Remove all temporary log files in pg_wal
3988 : *
3989 : * This is called at the beginning of recovery after a previous crash,
3990 : * at a point where no other processes write fresh WAL data.
3991 : */
3992 : static void
3993 176 : RemoveTempXlogFiles(void)
3994 : {
3995 : DIR *xldir;
3996 : struct dirent *xlde;
3997 :
3998 176 : elog(DEBUG2, "removing all temporary WAL segments");
3999 :
4000 176 : xldir = AllocateDir(XLOGDIR);
4001 966 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4002 : {
4003 : char path[MAXPGPATH];
4004 :
4005 790 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
4006 790 : continue;
4007 :
4008 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
4009 0 : unlink(path);
4010 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
4011 : }
4012 176 : FreeDir(xldir);
4013 176 : }
4014 :
4015 : /*
4016 : * Recycle or remove all log files older or equal to passed segno.
4017 : *
4018 : * endptr is current (or recent) end of xlog, and lastredoptr is the
4019 : * redo pointer of the last checkpoint. These are used to determine
4020 : * whether we want to recycle rather than delete no-longer-wanted log files.
4021 : */
4022 : static void
4023 3172 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr)
4024 : {
4025 : DIR *xldir;
4026 : struct dirent *xlde;
4027 : char lastoff[MAXFNAMELEN];
4028 :
4029 : /*
4030 : * Construct a filename of the last segment to be kept. The timeline ID
4031 : * doesn't matter, we ignore that in the comparison. (During recovery,
4032 : * ThisTimeLineID isn't set, so we can't use that.)
4033 : */
4034 3172 : XLogFileName(lastoff, 0, segno, wal_segment_size);
4035 :
4036 3172 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
4037 : lastoff);
4038 :
4039 3172 : xldir = AllocateDir(XLOGDIR);
4040 :
4041 17806 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4042 : {
4043 : /* Ignore files that are not XLOG segments */
4044 14634 : if (!IsXLogFileName(xlde->d_name) &&
4045 9602 : !IsPartialXLogFileName(xlde->d_name))
4046 9602 : continue;
4047 :
4048 : /*
4049 : * We ignore the timeline part of the XLOG segment identifiers in
4050 : * deciding whether a segment is still needed. This ensures that we
4051 : * won't prematurely remove a segment from a parent timeline. We could
4052 : * probably be a little more proactive about removing segments of
4053 : * non-parent timelines, but that would be a whole lot more
4054 : * complicated.
4055 : *
4056 : * We use the alphanumeric sorting property of the filenames to decide
4057 : * which ones are earlier than the lastoff segment.
4058 : */
4059 5032 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
4060 : {
4061 720 : if (XLogArchiveCheckDone(xlde->d_name))
4062 : {
4063 : /* Update the last removed location in shared memory first */
4064 718 : UpdateLastRemovedPtr(xlde->d_name);
4065 :
4066 718 : RemoveXlogFile(xlde->d_name, lastredoptr, endptr);
4067 : }
4068 : }
4069 : }
4070 :
4071 3172 : FreeDir(xldir);
4072 3172 : }
4073 :
4074 : /*
4075 : * Remove WAL files that are not part of the given timeline's history.
4076 : *
4077 : * This is called during recovery, whenever we switch to follow a new
4078 : * timeline, and at the end of recovery when we create a new timeline. We
4079 : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
4080 : * might be leftover pre-allocated or recycled WAL segments on the old timeline
4081 : * that we haven't used yet, and contain garbage. If we just leave them in
4082 : * pg_wal, they will eventually be archived, and we can't let that happen.
4083 : * Files that belong to our timeline history are valid, because we have
4084 : * successfully replayed them, but from others we can't be sure.
4085 : *
4086 : * 'switchpoint' is the current point in WAL where we switch to new timeline,
4087 : * and 'newTLI' is the new timeline we switch to.
4088 : */
4089 : static void
4090 58 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
4091 : {
4092 : DIR *xldir;
4093 : struct dirent *xlde;
4094 : char switchseg[MAXFNAMELEN];
4095 : XLogSegNo endLogSegNo;
4096 :
4097 58 : XLByteToPrevSeg(switchpoint, endLogSegNo, wal_segment_size);
4098 :
4099 : /*
4100 : * Construct a filename of the last segment to be kept.
4101 : */
4102 58 : XLogFileName(switchseg, newTLI, endLogSegNo, wal_segment_size);
4103 :
4104 58 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
4105 : switchseg);
4106 :
4107 58 : xldir = AllocateDir(XLOGDIR);
4108 :
4109 492 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4110 : {
4111 : /* Ignore files that are not XLOG segments */
4112 434 : if (!IsXLogFileName(xlde->d_name))
4113 250 : continue;
4114 :
4115 : /*
4116 : * Remove files that are on a timeline older than the new one we're
4117 : * switching to, but with a segment number >= the first segment on the
4118 : * new timeline.
4119 : */
4120 184 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
4121 122 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
4122 : {
4123 : /*
4124 : * If the file has already been marked as .ready, however, don't
4125 : * remove it yet. It should be OK to remove it - files that are
4126 : * not part of our timeline history are not required for recovery
4127 : * - but seems safer to let them be archived and removed later.
4128 : */
4129 22 : if (!XLogArchiveIsReady(xlde->d_name))
4130 22 : RemoveXlogFile(xlde->d_name, InvalidXLogRecPtr, switchpoint);
4131 : }
4132 : }
4133 :
4134 58 : FreeDir(xldir);
4135 58 : }
4136 :
4137 : /*
4138 : * Recycle or remove a log file that's no longer needed.
4139 : *
4140 : * endptr is current (or recent) end of xlog, and lastredoptr is the
4141 : * redo pointer of the last checkpoint. These are used to determine
4142 : * whether we want to recycle rather than delete no-longer-wanted log files.
4143 : * If lastredoptr is not known, pass invalid, and the function will recycle,
4144 : * somewhat arbitrarily, 10 future segments.
4145 : */
4146 : static void
4147 740 : RemoveXlogFile(const char *segname, XLogRecPtr lastredoptr, XLogRecPtr endptr)
4148 : {
4149 : char path[MAXPGPATH];
4150 : #ifdef WIN32
4151 : char newpath[MAXPGPATH];
4152 : #endif
4153 : struct stat statbuf;
4154 : XLogSegNo endlogSegNo;
4155 : XLogSegNo recycleSegNo;
4156 :
4157 740 : if (wal_recycle)
4158 : {
4159 : /*
4160 : * Initialize info about where to try to recycle to.
4161 : */
4162 740 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
4163 740 : if (lastredoptr == InvalidXLogRecPtr)
4164 22 : recycleSegNo = endlogSegNo + 10;
4165 : else
4166 718 : recycleSegNo = XLOGfileslop(lastredoptr);
4167 : }
4168 : else
4169 0 : recycleSegNo = 0; /* keep compiler quiet */
4170 :
4171 740 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4172 :
4173 : /*
4174 : * Before deleting the file, see if it can be recycled as a future log
4175 : * segment. Only recycle normal files, pg_standby for example can create
4176 : * symbolic links pointing to a separate archive directory.
4177 : */
4178 740 : if (wal_recycle &&
4179 1472 : endlogSegNo <= recycleSegNo &&
4180 2196 : lstat(path, &statbuf) == 0 && S_ISREG(statbuf.st_mode) &&
4181 732 : InstallXLogFileSegment(&endlogSegNo, path,
4182 : true, recycleSegNo, true))
4183 : {
4184 712 : ereport(DEBUG2,
4185 : (errmsg("recycled write-ahead log file \"%s\"",
4186 : segname)));
4187 712 : CheckpointStats.ckpt_segs_recycled++;
4188 : /* Needn't recheck that slot on future iterations */
4189 712 : endlogSegNo++;
4190 : }
4191 : else
4192 : {
4193 : /* No need for any more future segments... */
4194 : int rc;
4195 :
4196 28 : ereport(DEBUG2,
4197 : (errmsg("removing write-ahead log file \"%s\"",
4198 : segname)));
4199 :
4200 : #ifdef WIN32
4201 :
4202 : /*
4203 : * On Windows, if another process (e.g another backend) holds the file
4204 : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4205 : * will still show up in directory listing until the last handle is
4206 : * closed. To avoid confusing the lingering deleted file for a live
4207 : * WAL file that needs to be archived, rename it before deleting it.
4208 : *
4209 : * If another process holds the file open without FILE_SHARE_DELETE
4210 : * flag, rename will fail. We'll try again at the next checkpoint.
4211 : */
4212 : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4213 : if (rename(path, newpath) != 0)
4214 : {
4215 : ereport(LOG,
4216 : (errcode_for_file_access(),
4217 : errmsg("could not rename file \"%s\": %m",
4218 : path)));
4219 : return;
4220 : }
4221 : rc = durable_unlink(newpath, LOG);
4222 : #else
4223 28 : rc = durable_unlink(path, LOG);
4224 : #endif
4225 28 : if (rc != 0)
4226 : {
4227 : /* Message already logged by durable_unlink() */
4228 0 : return;
4229 : }
4230 28 : CheckpointStats.ckpt_segs_removed++;
4231 : }
4232 :
4233 740 : XLogArchiveCleanup(segname);
4234 : }
4235 :
4236 : /*
4237 : * Verify whether pg_wal and pg_wal/archive_status exist.
4238 : * If the latter does not exist, recreate it.
4239 : *
4240 : * It is not the goal of this function to verify the contents of these
4241 : * directories, but to help in cases where someone has performed a cluster
4242 : * copy for PITR purposes but omitted pg_wal from the copy.
4243 : *
4244 : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4245 : * policy decision was made not to. It is fairly common for pg_wal to be
4246 : * a symlink, and if that was the DBA's intent then automatically making a
4247 : * plain directory would result in degraded performance with no notice.
4248 : */
4249 : static void
4250 1390 : ValidateXLOGDirectoryStructure(void)
4251 : {
4252 : char path[MAXPGPATH];
4253 : struct stat stat_buf;
4254 :
4255 : /* Check for pg_wal; if it doesn't exist, error out */
4256 1390 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4257 1390 : !S_ISDIR(stat_buf.st_mode))
4258 0 : ereport(FATAL,
4259 : (errmsg("required WAL directory \"%s\" does not exist",
4260 : XLOGDIR)));
4261 :
4262 : /* Check for archive_status */
4263 1390 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4264 1390 : if (stat(path, &stat_buf) == 0)
4265 : {
4266 : /* Check for weird cases where it exists but isn't a directory */
4267 1390 : if (!S_ISDIR(stat_buf.st_mode))
4268 0 : ereport(FATAL,
4269 : (errmsg("required WAL directory \"%s\" does not exist",
4270 : path)));
4271 : }
4272 : else
4273 : {
4274 0 : ereport(LOG,
4275 : (errmsg("creating missing WAL directory \"%s\"", path)));
4276 0 : if (MakePGDirectory(path) < 0)
4277 0 : ereport(FATAL,
4278 : (errmsg("could not create missing directory \"%s\": %m",
4279 : path)));
4280 : }
4281 1390 : }
4282 :
4283 : /*
4284 : * Remove previous backup history files. This also retries creation of
4285 : * .ready files for any backup history files for which XLogArchiveNotify
4286 : * failed earlier.
4287 : */
4288 : static void
4289 136 : CleanupBackupHistory(void)
4290 : {
4291 : DIR *xldir;
4292 : struct dirent *xlde;
4293 : char path[MAXPGPATH + sizeof(XLOGDIR)];
4294 :
4295 136 : xldir = AllocateDir(XLOGDIR);
4296 :
4297 1040 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4298 : {
4299 904 : if (IsBackupHistoryFileName(xlde->d_name))
4300 : {
4301 136 : if (XLogArchiveCheckDone(xlde->d_name))
4302 : {
4303 130 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4304 : xlde->d_name);
4305 130 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
4306 130 : unlink(path);
4307 130 : XLogArchiveCleanup(xlde->d_name);
4308 : }
4309 : }
4310 : }
4311 :
4312 136 : FreeDir(xldir);
4313 136 : }
4314 :
4315 : /*
4316 : * Attempt to read the next XLOG record.
4317 : *
4318 : * Before first call, the reader needs to be positioned to the first record
4319 : * by calling XLogBeginRead().
4320 : *
4321 : * If no valid record is available, returns NULL, or fails if emode is PANIC.
4322 : * (emode must be either PANIC, LOG). In standby mode, retries until a valid
4323 : * record is available.
4324 : */
4325 : static XLogRecord *
4326 576502 : ReadRecord(XLogReaderState *xlogreader, int emode,
4327 : bool fetching_ckpt)
4328 : {
4329 : XLogRecord *record;
4330 576502 : XLogPageReadPrivate *private = (XLogPageReadPrivate *) xlogreader->private_data;
4331 :
4332 : /* Pass through parameters to XLogPageRead */
4333 576502 : private->fetching_ckpt = fetching_ckpt;
4334 576502 : private->emode = emode;
4335 576502 : private->randAccess = (xlogreader->ReadRecPtr == InvalidXLogRecPtr);
4336 :
4337 : /* This is the first attempt to read this page. */
4338 576502 : lastSourceFailed = false;
4339 :
4340 : for (;;)
4341 92 : {
4342 : char *errormsg;
4343 :
4344 576594 : record = XLogReadRecord(xlogreader, &errormsg);
4345 576562 : ReadRecPtr = xlogreader->ReadRecPtr;
4346 576562 : EndRecPtr = xlogreader->EndRecPtr;
4347 576562 : if (record == NULL)
4348 : {
4349 256 : if (readFile >= 0)
4350 : {
4351 232 : close(readFile);
4352 232 : readFile = -1;
4353 : }
4354 :
4355 : /*
4356 : * We only end up here without a message when XLogPageRead()
4357 : * failed - in that case we already logged something. In
4358 : * StandbyMode that only happens if we have been triggered, so we
4359 : * shouldn't loop anymore in that case.
4360 : */
4361 256 : if (errormsg)
4362 232 : ereport(emode_for_corrupt_record(emode, EndRecPtr),
4363 : (errmsg_internal("%s", errormsg) /* already translated */ ));
4364 : }
4365 :
4366 : /*
4367 : * Check page TLI is one of the expected values.
4368 : */
4369 576306 : else if (!tliInHistory(xlogreader->latestPageTLI, expectedTLEs))
4370 : {
4371 : char fname[MAXFNAMELEN];
4372 : XLogSegNo segno;
4373 : int32 offset;
4374 :
4375 0 : XLByteToSeg(xlogreader->latestPagePtr, segno, wal_segment_size);
4376 0 : offset = XLogSegmentOffset(xlogreader->latestPagePtr,
4377 : wal_segment_size);
4378 0 : XLogFileName(fname, xlogreader->seg.ws_tli, segno,
4379 : wal_segment_size);
4380 0 : ereport(emode_for_corrupt_record(emode, EndRecPtr),
4381 : (errmsg("unexpected timeline ID %u in log segment %s, offset %u",
4382 : xlogreader->latestPageTLI,
4383 : fname,
4384 : offset)));
4385 0 : record = NULL;
4386 : }
4387 :
4388 576562 : if (record)
4389 : {
4390 : /* Great, got a record */
4391 576470 : return record;
4392 : }
4393 : else
4394 : {
4395 : /* No valid record available from this source */
4396 256 : lastSourceFailed = true;
4397 :
4398 : /*
4399 : * If archive recovery was requested, but we were still doing
4400 : * crash recovery, switch to archive recovery and retry using the
4401 : * offline archive. We have now replayed all the valid WAL in
4402 : * pg_wal, so we are presumably now consistent.
4403 : *
4404 : * We require that there's at least some valid WAL present in
4405 : * pg_wal, however (!fetching_ckpt). We could recover using the
4406 : * WAL from the archive, even if pg_wal is completely empty, but
4407 : * we'd have no idea how far we'd have to replay to reach
4408 : * consistency. So err on the safe side and give up.
4409 : */
4410 256 : if (!InArchiveRecovery && ArchiveRecoveryRequested &&
4411 2 : !fetching_ckpt)
4412 : {
4413 2 : ereport(DEBUG1,
4414 : (errmsg_internal("reached end of WAL in pg_wal, entering archive recovery")));
4415 2 : InArchiveRecovery = true;
4416 2 : if (StandbyModeRequested)
4417 2 : StandbyMode = true;
4418 :
4419 : /* initialize minRecoveryPoint to this record */
4420 2 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
4421 2 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
4422 2 : if (ControlFile->minRecoveryPoint < EndRecPtr)
4423 : {
4424 2 : ControlFile->minRecoveryPoint = EndRecPtr;
4425 2 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
4426 : }
4427 : /* update local copy */
4428 2 : minRecoveryPoint = ControlFile->minRecoveryPoint;
4429 2 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
4430 :
4431 : /*
4432 : * The startup process can update its local copy of
4433 : * minRecoveryPoint from this point.
4434 : */
4435 2 : updateMinRecoveryPoint = true;
4436 :
4437 2 : UpdateControlFile();
4438 :
4439 : /*
4440 : * We update SharedRecoveryState while holding the lock on
4441 : * ControlFileLock so both states are consistent in shared
4442 : * memory.
4443 : */
4444 2 : SpinLockAcquire(&XLogCtl->info_lck);
4445 2 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
4446 2 : SpinLockRelease(&XLogCtl->info_lck);
4447 :
4448 2 : LWLockRelease(ControlFileLock);
4449 :
4450 2 : CheckRecoveryConsistency();
4451 :
4452 : /*
4453 : * Before we retry, reset lastSourceFailed and currentSource
4454 : * so that we will check the archive next.
4455 : */
4456 2 : lastSourceFailed = false;
4457 2 : currentSource = XLOG_FROM_ANY;
4458 :
4459 92 : continue;
4460 : }
4461 :
4462 : /* In standby mode, loop back to retry. Otherwise, give up. */
4463 254 : if (StandbyMode && !CheckForStandbyTrigger())
4464 90 : continue;
4465 : else
4466 164 : return NULL;
4467 : }
4468 : }
4469 : }
4470 :
4471 : /*
4472 : * Scan for new timelines that might have appeared in the archive since we
4473 : * started recovery.
4474 : *
4475 : * If there are any, the function changes recovery target TLI to the latest
4476 : * one and returns 'true'.
4477 : */
4478 : static bool
4479 142 : rescanLatestTimeLine(void)
4480 : {
4481 : List *newExpectedTLEs;
4482 : bool found;
4483 : ListCell *cell;
4484 : TimeLineID newtarget;
4485 142 : TimeLineID oldtarget = recoveryTargetTLI;
4486 142 : TimeLineHistoryEntry *currentTle = NULL;
4487 :
4488 142 : newtarget = findNewestTimeLine(recoveryTargetTLI);
4489 142 : if (newtarget == recoveryTargetTLI)
4490 : {
4491 : /* No new timelines found */
4492 136 : return false;
4493 : }
4494 :
4495 : /*
4496 : * Determine the list of expected TLIs for the new TLI
4497 : */
4498 :
4499 6 : newExpectedTLEs = readTimeLineHistory(newtarget);
4500 :
4501 : /*
4502 : * If the current timeline is not part of the history of the new timeline,
4503 : * we cannot proceed to it.
4504 : */
4505 6 : found = false;
4506 12 : foreach(cell, newExpectedTLEs)
4507 : {
4508 12 : currentTle = (TimeLineHistoryEntry *) lfirst(cell);
4509 :
4510 12 : if (currentTle->tli == recoveryTargetTLI)
4511 : {
4512 6 : found = true;
4513 6 : break;
4514 : }
4515 : }
4516 6 : if (!found)
4517 : {
4518 0 : ereport(LOG,
4519 : (errmsg("new timeline %u is not a child of database system timeline %u",
4520 : newtarget,
4521 : ThisTimeLineID)));
4522 0 : return false;
4523 : }
4524 :
4525 : /*
4526 : * The current timeline was found in the history file, but check that the
4527 : * next timeline was forked off from it *after* the current recovery
4528 : * location.
4529 : */
4530 6 : if (currentTle->end < EndRecPtr)
4531 : {
4532 0 : ereport(LOG,
4533 : (errmsg("new timeline %u forked off current database system timeline %u before current recovery point %X/%X",
4534 : newtarget,
4535 : ThisTimeLineID,
4536 : (uint32) (EndRecPtr >> 32), (uint32) EndRecPtr)));
4537 0 : return false;
4538 : }
4539 :
4540 : /* The new timeline history seems valid. Switch target */
4541 6 : recoveryTargetTLI = newtarget;
4542 6 : list_free_deep(expectedTLEs);
4543 6 : expectedTLEs = newExpectedTLEs;
4544 :
4545 : /*
4546 : * As in StartupXLOG(), try to ensure we have all the history files
4547 : * between the old target and new target in pg_wal.
4548 : */
4549 6 : restoreTimeLineHistoryFiles(oldtarget + 1, newtarget);
4550 :
4551 6 : ereport(LOG,
4552 : (errmsg("new target timeline is %u",
4553 : recoveryTargetTLI)));
4554 :
4555 6 : return true;
4556 : }
4557 :
4558 : /*
4559 : * I/O routines for pg_control
4560 : *
4561 : * *ControlFile is a buffer in shared memory that holds an image of the
4562 : * contents of pg_control. WriteControlFile() initializes pg_control
4563 : * given a preloaded buffer, ReadControlFile() loads the buffer from
4564 : * the pg_control file (during postmaster or standalone-backend startup),
4565 : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4566 : * InitControlFile() fills the buffer with initial values.
4567 : *
4568 : * For simplicity, WriteControlFile() initializes the fields of pg_control
4569 : * that are related to checking backend/database compatibility, and
4570 : * ReadControlFile() verifies they are correct. We could split out the
4571 : * I/O and compatibility-check functions, but there seems no need currently.
4572 : */
4573 :
4574 : static void
4575 358 : InitControlFile(uint64 sysidentifier)
4576 : {
4577 : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4578 :
4579 : /*
4580 : * Generate a random nonce. This is used for authentication requests that
4581 : * will fail because the user does not exist. The nonce is used to create
4582 : * a genuine-looking password challenge for the non-existent user, in lieu
4583 : * of an actual stored password.
4584 : */
4585 358 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
4586 0 : ereport(PANIC,
4587 : (errcode(ERRCODE_INTERNAL_ERROR),
4588 : errmsg("could not generate secret authorization token")));
4589 :
4590 358 : memset(ControlFile, 0, sizeof(ControlFileData));
4591 : /* Initialize pg_control status fields */
4592 358 : ControlFile->system_identifier = sysidentifier;
4593 358 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4594 358 : ControlFile->state = DB_SHUTDOWNED;
4595 358 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
4596 :
4597 : /* Set important parameter values for use when replaying WAL */
4598 358 : ControlFile->MaxConnections = MaxConnections;
4599 358 : ControlFile->max_worker_processes = max_worker_processes;
4600 358 : ControlFile->max_wal_senders = max_wal_senders;
4601 358 : ControlFile->max_prepared_xacts = max_prepared_xacts;
4602 358 : ControlFile->max_locks_per_xact = max_locks_per_xact;
4603 358 : ControlFile->wal_level = wal_level;
4604 358 : ControlFile->wal_log_hints = wal_log_hints;
4605 358 : ControlFile->track_commit_timestamp = track_commit_timestamp;
4606 358 : ControlFile->data_checksum_version = bootstrap_data_checksum_version;
4607 358 : }
4608 :
4609 : static void
4610 358 : WriteControlFile(void)
4611 : {
4612 : int fd;
4613 : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4614 :
4615 : /*
4616 : * Ensure that the size of the pg_control data structure is sane. See the
4617 : * comments for these symbols in pg_control.h.
4618 : */
4619 : StaticAssertStmt(sizeof(ControlFileData) <= PG_CONTROL_MAX_SAFE_SIZE,
4620 : "pg_control is too large for atomic disk writes");
4621 : StaticAssertStmt(sizeof(ControlFileData) <= PG_CONTROL_FILE_SIZE,
4622 : "sizeof(ControlFileData) exceeds PG_CONTROL_FILE_SIZE");
4623 :
4624 : /*
4625 : * Initialize version and compatibility-check fields
4626 : */
4627 358 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4628 358 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4629 :
4630 358 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4631 358 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4632 :
4633 358 : ControlFile->blcksz = BLCKSZ;
4634 358 : ControlFile->relseg_size = RELSEG_SIZE;
4635 358 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
4636 358 : ControlFile->xlog_seg_size = wal_segment_size;
4637 :
4638 358 : ControlFile->nameDataLen = NAMEDATALEN;
4639 358 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4640 :
4641 358 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4642 358 : ControlFile->loblksize = LOBLKSIZE;
4643 :
4644 358 : ControlFile->float8ByVal = FLOAT8PASSBYVAL;
4645 :
4646 : /* Contents are protected with a CRC */
4647 358 : INIT_CRC32C(ControlFile->crc);
4648 358 : COMP_CRC32C(ControlFile->crc,
4649 : (char *) ControlFile,
4650 : offsetof(ControlFileData, crc));
4651 358 : FIN_CRC32C(ControlFile->crc);
4652 :
4653 : /*
4654 : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4655 : * the excess over sizeof(ControlFileData). This reduces the odds of
4656 : * premature-EOF errors when reading pg_control. We'll still fail when we
4657 : * check the contents of the file, but hopefully with a more specific
4658 : * error than "couldn't read pg_control".
4659 : */
4660 358 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
4661 358 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4662 :
4663 358 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4664 : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
4665 358 : if (fd < 0)
4666 0 : ereport(PANIC,
4667 : (errcode_for_file_access(),
4668 : errmsg("could not create file \"%s\": %m",
4669 : XLOG_CONTROL_FILE)));
4670 :
4671 358 : errno = 0;
4672 358 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
4673 358 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4674 : {
4675 : /* if write didn't set errno, assume problem is no disk space */
4676 0 : if (errno == 0)
4677 0 : errno = ENOSPC;
4678 0 : ereport(PANIC,
4679 : (errcode_for_file_access(),
4680 : errmsg("could not write to file \"%s\": %m",
4681 : XLOG_CONTROL_FILE)));
4682 : }
4683 358 : pgstat_report_wait_end();
4684 :
4685 358 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
4686 358 : if (pg_fsync(fd) != 0)
4687 0 : ereport(PANIC,
4688 : (errcode_for_file_access(),
4689 : errmsg("could not fsync file \"%s\": %m",
4690 : XLOG_CONTROL_FILE)));
4691 358 : pgstat_report_wait_end();
4692 :
4693 358 : if (close(fd) != 0)
4694 0 : ereport(PANIC,
4695 : (errcode_for_file_access(),
4696 : errmsg("could not close file \"%s\": %m",
4697 : XLOG_CONTROL_FILE)));
4698 358 : }
4699 :
4700 : static void
4701 1464 : ReadControlFile(void)
4702 : {
4703 : pg_crc32c crc;
4704 : int fd;
4705 : static char wal_segsz_str[20];
4706 : int r;
4707 :
4708 : /*
4709 : * Read data...
4710 : */
4711 1464 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4712 : O_RDWR | PG_BINARY);
4713 1464 : if (fd < 0)
4714 0 : ereport(PANIC,
4715 : (errcode_for_file_access(),
4716 : errmsg("could not open file \"%s\": %m",
4717 : XLOG_CONTROL_FILE)));
4718 :
4719 1464 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
4720 1464 : r = read(fd, ControlFile, sizeof(ControlFileData));
4721 1464 : if (r != sizeof(ControlFileData))
4722 : {
4723 0 : if (r < 0)
4724 0 : ereport(PANIC,
4725 : (errcode_for_file_access(),
4726 : errmsg("could not read file \"%s\": %m",
4727 : XLOG_CONTROL_FILE)));
4728 : else
4729 0 : ereport(PANIC,
4730 : (errcode(ERRCODE_DATA_CORRUPTED),
4731 : errmsg("could not read file \"%s\": read %d of %zu",
4732 : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4733 : }
4734 1464 : pgstat_report_wait_end();
4735 :
4736 1464 : close(fd);
4737 :
4738 : /*
4739 : * Check for expected pg_control format version. If this is wrong, the
4740 : * CRC check will likely fail because we'll be checking the wrong number
4741 : * of bytes. Complaining about wrong version will probably be more
4742 : * enlightening than complaining about wrong CRC.
4743 : */
4744 :
4745 1464 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
4746 0 : ereport(FATAL,
4747 : (errmsg("database files are incompatible with server"),
4748 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4749 : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4750 : ControlFile->pg_control_version, ControlFile->pg_control_version,
4751 : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4752 : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4753 :
4754 1464 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
4755 0 : ereport(FATAL,
4756 : (errmsg("database files are incompatible with server"),
4757 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4758 : " but the server was compiled with PG_CONTROL_VERSION %d.",
4759 : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4760 : errhint("It looks like you need to initdb.")));
4761 :
4762 : /* Now check the CRC. */
4763 1464 : INIT_CRC32C(crc);
4764 1464 : COMP_CRC32C(crc,
4765 : (char *) ControlFile,
4766 : offsetof(ControlFileData, crc));
4767 1464 : FIN_CRC32C(crc);
4768 :
4769 1464 : if (!EQ_CRC32C(crc, ControlFile->crc))
4770 0 : ereport(FATAL,
4771 : (errmsg("incorrect checksum in control file")));
4772 :
4773 : /*
4774 : * Do compatibility checking immediately. If the database isn't
4775 : * compatible with the backend executable, we want to abort before we can
4776 : * possibly do any damage.
4777 : */
4778 1464 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
4779 0 : ereport(FATAL,
4780 : (errmsg("database files are incompatible with server"),
4781 : errdetail("The database cluster was initialized with CATALOG_VERSION_NO %d,"
4782 : " but the server was compiled with CATALOG_VERSION_NO %d.",
4783 : ControlFile->catalog_version_no, CATALOG_VERSION_NO),
4784 : errhint("It looks like you need to initdb.")));
4785 1464 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
4786 0 : ereport(FATAL,
4787 : (errmsg("database files are incompatible with server"),
4788 : errdetail("The database cluster was initialized with MAXALIGN %d,"
4789 : " but the server was compiled with MAXALIGN %d.",
4790 : ControlFile->maxAlign, MAXIMUM_ALIGNOF),
4791 : errhint("It looks like you need to initdb.")));
4792 1464 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
4793 0 : ereport(FATAL,
4794 : (errmsg("database files are incompatible with server"),
4795 : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4796 : errhint("It looks like you need to initdb.")));
4797 1464 : if (ControlFile->blcksz != BLCKSZ)
4798 0 : ereport(FATAL,
4799 : (errmsg("database files are incompatible with server"),
4800 : errdetail("The database cluster was initialized with BLCKSZ %d,"
4801 : " but the server was compiled with BLCKSZ %d.",
4802 : ControlFile->blcksz, BLCKSZ),
4803 : errhint("It looks like you need to recompile or initdb.")));
4804 1464 : if (ControlFile->relseg_size != RELSEG_SIZE)
4805 0 : ereport(FATAL,
4806 : (errmsg("database files are incompatible with server"),
4807 : errdetail("The database cluster was initialized with RELSEG_SIZE %d,"
4808 : " but the server was compiled with RELSEG_SIZE %d.",
4809 : ControlFile->relseg_size, RELSEG_SIZE),
4810 : errhint("It looks like you need to recompile or initdb.")));
4811 1464 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
4812 0 : ereport(FATAL,
4813 : (errmsg("database files are incompatible with server"),
4814 : errdetail("The database cluster was initialized with XLOG_BLCKSZ %d,"
4815 : " but the server was compiled with XLOG_BLCKSZ %d.",
4816 : ControlFile->xlog_blcksz, XLOG_BLCKSZ),
4817 : errhint("It looks like you need to recompile or initdb.")));
4818 1464 : if (ControlFile->nameDataLen != NAMEDATALEN)
4819 0 : ereport(FATAL,
4820 : (errmsg("database files are incompatible with server"),
4821 : errdetail("The database cluster was initialized with NAMEDATALEN %d,"
4822 : " but the server was compiled with NAMEDATALEN %d.",
4823 : ControlFile->nameDataLen, NAMEDATALEN),
4824 : errhint("It looks like you need to recompile or initdb.")));
4825 1464 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
4826 0 : ereport(FATAL,
4827 : (errmsg("database files are incompatible with server"),
4828 : errdetail("The database cluster was initialized with INDEX_MAX_KEYS %d,"
4829 : " but the server was compiled with INDEX_MAX_KEYS %d.",
4830 : ControlFile->indexMaxKeys, INDEX_MAX_KEYS),
4831 : errhint("It looks like you need to recompile or initdb.")));
4832 1464 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
4833 0 : ereport(FATAL,
4834 : (errmsg("database files are incompatible with server"),
4835 : errdetail("The database cluster was initialized with TOAST_MAX_CHUNK_SIZE %d,"
4836 : " but the server was compiled with TOAST_MAX_CHUNK_SIZE %d.",
4837 : ControlFile->toast_max_chunk_size, (int) TOAST_MAX_CHUNK_SIZE),
4838 : errhint("It looks like you need to recompile or initdb.")));
4839 1464 : if (ControlFile->loblksize != LOBLKSIZE)
4840 0 : ereport(FATAL,
4841 : (errmsg("database files are incompatible with server"),
4842 : errdetail("The database cluster was initialized with LOBLKSIZE %d,"
4843 : " but the server was compiled with LOBLKSIZE %d.",
4844 : ControlFile->loblksize, (int) LOBLKSIZE),
4845 : errhint("It looks like you need to recompile or initdb.")));
4846 :
4847 : #ifdef USE_FLOAT8_BYVAL
4848 1464 : if (ControlFile->float8ByVal != true)
4849 0 : ereport(FATAL,
4850 : (errmsg("database files are incompatible with server"),
4851 : errdetail("The database cluster was initialized without USE_FLOAT8_BYVAL"
4852 : " but the server was compiled with USE_FLOAT8_BYVAL."),
4853 : errhint("It looks like you need to recompile or initdb.")));
4854 : #else
4855 : if (ControlFile->float8ByVal != false)
4856 : ereport(FATAL,
4857 : (errmsg("database files are incompatible with server"),
4858 : errdetail("The database cluster was initialized with USE_FLOAT8_BYVAL"
4859 : " but the server was compiled without USE_FLOAT8_BYVAL."),
4860 : errhint("It looks like you need to recompile or initdb.")));
4861 : #endif
4862 :
4863 1464 : wal_segment_size = ControlFile->xlog_seg_size;
4864 :
4865 1464 : if (!IsValidWalSegSize(wal_segment_size))
4866 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4867 : errmsg_plural("WAL segment size must be a power of two between 1 MB and 1 GB, but the control file specifies %d byte",
4868 : "WAL segment size must be a power of two between 1 MB and 1 GB, but the control file specifies %d bytes",
4869 : wal_segment_size,
4870 : wal_segment_size)));
4871 :
4872 1464 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4873 1464 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4874 : PGC_S_OVERRIDE);
4875 :
4876 : /* check and update variables dependent on wal_segment_size */
4877 1464 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
4878 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4879 : errmsg("\"min_wal_size\" must be at least twice \"wal_segment_size\"")));
4880 :
4881 1464 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
4882 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4883 : errmsg("\"max_wal_size\" must be at least twice \"wal_segment_size\"")));
4884 :
4885 1464 : UsableBytesInSegment =
4886 1464 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4887 : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4888 :
4889 1464 : CalculateCheckpointSegments();
4890 :
4891 : /* Make the initdb settings visible as GUC variables, too */
4892 1464 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4893 : PGC_INTERNAL, PGC_S_OVERRIDE);
4894 1464 : }
4895 :
4896 : /*
4897 : * Utility wrapper to update the control file. Note that the control
4898 : * file gets flushed.
4899 : */
4900 : void
4901 6784 : UpdateControlFile(void)
4902 : {
4903 6784 : update_controlfile(DataDir, ControlFile, true);
4904 6784 : }
4905 :
4906 : /*
4907 : * Returns the unique system identifier from control file.
4908 : */
4909 : uint64
4910 592 : GetSystemIdentifier(void)
4911 : {
4912 : Assert(ControlFile != NULL);
4913 592 : return ControlFile->system_identifier;
4914 : }
4915 :
4916 : /*
4917 : * Returns the random nonce from control file.
4918 : */
4919 : char *
4920 2 : GetMockAuthenticationNonce(void)
4921 : {
4922 : Assert(ControlFile != NULL);
4923 2 : return ControlFile->mock_authentication_nonce;
4924 : }
4925 :
4926 : /*
4927 : * Are checksums enabled for data pages?
4928 : */
4929 : bool
4930 14926094 : DataChecksumsEnabled(void)
4931 : {
4932 : Assert(ControlFile != NULL);
4933 14926094 : return (ControlFile->data_checksum_version > 0);
4934 : }
4935 :
4936 : /*
4937 : * Returns a fake LSN for unlogged relations.
4938 : *
4939 : * Each call generates an LSN that is greater than any previous value
4940 : * returned. The current counter value is saved and restored across clean
4941 : * shutdowns, but like unlogged relations, does not survive a crash. This can
4942 : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4943 : * LSN-like increasing sequence of numbers without writing any WAL.
4944 : */
4945 : XLogRecPtr
4946 0 : GetFakeLSNForUnloggedRel(void)
4947 : {
4948 : XLogRecPtr nextUnloggedLSN;
4949 :
4950 : /* increment the unloggedLSN counter, need SpinLock */
4951 0 : SpinLockAcquire(&XLogCtl->ulsn_lck);
4952 0 : nextUnloggedLSN = XLogCtl->unloggedLSN++;
4953 0 : SpinLockRelease(&XLogCtl->ulsn_lck);
4954 :
4955 0 : return nextUnloggedLSN;
4956 : }
4957 :
4958 : /*
4959 : * Auto-tune the number of XLOG buffers.
4960 : *
4961 : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4962 : * a maximum of one XLOG segment (there is little reason to think that more
4963 : * is helpful, at least so long as we force an fsync when switching log files)
4964 : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4965 : * 9.1, when auto-tuning was added).
4966 : *
4967 : * This should not be called until NBuffers has received its final value.
4968 : */
4969 : static int
4970 2170 : XLOGChooseNumBuffers(void)
4971 : {
4972 : int xbuffers;
4973 :
4974 2170 : xbuffers = NBuffers / 32;
4975 2170 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4976 24 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
4977 2170 : if (xbuffers < 8)
4978 358 : xbuffers = 8;
4979 2170 : return xbuffers;
4980 : }
4981 :
4982 : /*
4983 : * GUC check_hook for wal_buffers
4984 : */
4985 : bool
4986 4354 : check_wal_buffers(int *newval, void **extra, GucSource source)
4987 : {
4988 : /*
4989 : * -1 indicates a request for auto-tune.
4990 : */
4991 4354 : if (*newval == -1)
4992 : {
4993 : /*
4994 : * If we haven't yet changed the boot_val default of -1, just let it
4995 : * be. We'll fix it when XLOGShmemSize is called.
4996 : */
4997 2184 : if (XLOGbuffers == -1)
4998 2184 : return true;
4999 :
5000 : /* Otherwise, substitute the auto-tune value */
5001 0 : *newval = XLOGChooseNumBuffers();
5002 : }
5003 :
5004 : /*
5005 : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
5006 : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
5007 : * the case, we just silently treat such values as a request for the
5008 : * minimum. (We could throw an error instead, but that doesn't seem very
5009 : * helpful.)
5010 : */
5011 2170 : if (*newval < 4)
5012 0 : *newval = 4;
5013 :
5014 2170 : return true;
5015 : }
5016 :
5017 : /*
5018 : * Read the control file, set respective GUCs.
5019 : *
5020 : * This is to be called during startup, including a crash recovery cycle,
5021 : * unless in bootstrap mode, where no control file yet exists. As there's no
5022 : * usable shared memory yet (its sizing can depend on the contents of the
5023 : * control file!), first store the contents in local memory. XLOGShmemInit()
5024 : * will then copy it to shared memory later.
5025 : *
5026 : * reset just controls whether previous contents are to be expected (in the
5027 : * reset case, there's a dangling pointer into old shared memory), or not.
5028 : */
5029 : void
5030 1106 : LocalProcessControlFile(bool reset)
5031 : {
5032 : Assert(reset || ControlFile == NULL);
5033 1106 : ControlFile = palloc(sizeof(ControlFileData));
5034 1106 : ReadControlFile();
5035 1106 : }
5036 :
5037 : /*
5038 : * Initialization of shared memory for XLOG
5039 : */
5040 : Size
5041 4344 : XLOGShmemSize(void)
5042 : {
5043 : Size size;
5044 :
5045 : /*
5046 : * If the value of wal_buffers is -1, use the preferred auto-tune value.
5047 : * This isn't an amazingly clean place to do this, but we must wait till
5048 : * NBuffers has received its final value, and must do it before using the
5049 : * value of XLOGbuffers to do anything important.
5050 : */
5051 4344 : if (XLOGbuffers == -1)
5052 : {
5053 : char buf[32];
5054 :
5055 2170 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
5056 2170 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER, PGC_S_OVERRIDE);
5057 : }
5058 : Assert(XLOGbuffers > 0);
5059 :
5060 : /* XLogCtl */
5061 4344 : size = sizeof(XLogCtlData);
5062 :
5063 : /* WAL insertion locks, plus alignment */
5064 4344 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
5065 : /* xlblocks array */
5066 4344 : size = add_size(size, mul_size(sizeof(XLogRecPtr), XLOGbuffers));
5067 : /* extra alignment padding for XLOG I/O buffers */
5068 4344 : size = add_size(size, XLOG_BLCKSZ);
5069 : /* and the buffers themselves */
5070 4344 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
5071 :
5072 : /*
5073 : * Note: we don't count ControlFileData, it comes out of the "slop factor"
5074 : * added by CreateSharedMemoryAndSemaphores. This lets us use this
5075 : * routine again below to compute the actual allocation size.
5076 : */
5077 :
5078 4344 : return size;
5079 : }
5080 :
5081 : void
5082 2170 : XLOGShmemInit(void)
5083 : {
5084 : bool foundCFile,
5085 : foundXLog;
5086 : char *allocptr;
5087 : int i;
5088 : ControlFileData *localControlFile;
5089 :
5090 : #ifdef WAL_DEBUG
5091 :
5092 : /*
5093 : * Create a memory context for WAL debugging that's exempt from the normal
5094 : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
5095 : * an allocation fails, but wal_debug is not for production use anyway.
5096 : */
5097 : if (walDebugCxt == NULL)
5098 : {
5099 : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
5100 : "WAL Debug",
5101 : ALLOCSET_DEFAULT_SIZES);
5102 : MemoryContextAllowInCriticalSection(walDebugCxt, true);
5103 : }
5104 : #endif
5105 :
5106 :
5107 2170 : XLogCtl = (XLogCtlData *)
5108 2170 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
5109 :
5110 2170 : localControlFile = ControlFile;
5111 2170 : ControlFile = (ControlFileData *)
5112 2170 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
5113 :
5114 2170 : if (foundCFile || foundXLog)
5115 : {
5116 : /* both should be present or neither */
5117 : Assert(foundCFile && foundXLog);
5118 :
5119 : /* Initialize local copy of WALInsertLocks */
5120 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
5121 :
5122 0 : if (localControlFile)
5123 0 : pfree(localControlFile);
5124 0 : return;
5125 : }
5126 2170 : memset(XLogCtl, 0, sizeof(XLogCtlData));
5127 :
5128 : /*
5129 : * Already have read control file locally, unless in bootstrap mode. Move
5130 : * contents into shared memory.
5131 : */
5132 2170 : if (localControlFile)
5133 : {
5134 1096 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
5135 1096 : pfree(localControlFile);
5136 : }
5137 :
5138 : /*
5139 : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
5140 : * multiple of the alignment for same, so no extra alignment padding is
5141 : * needed here.
5142 : */
5143 2170 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
5144 2170 : XLogCtl->xlblocks = (XLogRecPtr *) allocptr;
5145 2170 : memset(XLogCtl->xlblocks, 0, sizeof(XLogRecPtr) * XLOGbuffers);
5146 2170 : allocptr += sizeof(XLogRecPtr) * XLOGbuffers;
5147 :
5148 :
5149 : /* WAL insertion locks. Ensure they're aligned to the full padded size */
5150 2170 : allocptr += sizeof(WALInsertLockPadded) -
5151 2170 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
5152 2170 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
5153 : (WALInsertLockPadded *) allocptr;
5154 2170 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
5155 :
5156 19530 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
5157 : {
5158 17360 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
5159 17360 : WALInsertLocks[i].l.insertingAt = InvalidXLogRecPtr;
5160 17360 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
5161 : }
5162 :
5163 : /*
5164 : * Align the start of the page buffers to a full xlog block size boundary.
5165 : * This simplifies some calculations in XLOG insertion. It is also
5166 : * required for O_DIRECT.
5167 : */
5168 2170 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
5169 2170 : XLogCtl->pages = allocptr;
5170 2170 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5171 :
5172 : /*
5173 : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5174 : * in additional info.)
5175 : */
5176 2170 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
5177 2170 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5178 2170 : XLogCtl->SharedHotStandbyActive = false;
5179 2170 : XLogCtl->SharedPromoteIsTriggered = false;
5180 2170 : XLogCtl->WalWriterSleeping = false;
5181 :
5182 2170 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
5183 2170 : SpinLockInit(&XLogCtl->info_lck);
5184 2170 : SpinLockInit(&XLogCtl->ulsn_lck);
5185 2170 : InitSharedLatch(&XLogCtl->recoveryWakeupLatch);
5186 : }
5187 :
5188 : /*
5189 : * This func must be called ONCE on system install. It creates pg_control
5190 : * and the initial XLOG segment.
5191 : */
5192 : void
5193 358 : BootStrapXLOG(void)
5194 : {
5195 : CheckPoint checkPoint;
5196 : char *buffer;
5197 : XLogPageHeader page;
5198 : XLogLongPageHeader longpage;
5199 : XLogRecord *record;
5200 : char *recptr;
5201 : bool use_existent;
5202 : uint64 sysidentifier;
5203 : struct timeval tv;
5204 : pg_crc32c crc;
5205 :
5206 : /*
5207 : * Select a hopefully-unique system identifier code for this installation.
5208 : * We use the result of gettimeofday(), including the fractional seconds
5209 : * field, as being about as unique as we can easily get. (Think not to
5210 : * use random(), since it hasn't been seeded and there's no portable way
5211 : * to seed it other than the system clock value...) The upper half of the
5212 : * uint64 value is just the tv_sec part, while the lower half contains the
5213 : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5214 : * PID for a little extra uniqueness. A person knowing this encoding can
5215 : * determine the initialization time of the installation, which could
5216 : * perhaps be useful sometimes.
5217 : */
5218 358 : gettimeofday(&tv, NULL);
5219 358 : sysidentifier = ((uint64) tv.tv_sec) << 32;
5220 358 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5221 358 : sysidentifier |= getpid() & 0xFFF;
5222 :
5223 : /* First timeline ID is always 1 */
5224 358 : ThisTimeLineID = 1;
5225 :
5226 : /* page buffer must be aligned suitably for O_DIRECT */
5227 358 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5228 358 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
5229 358 : memset(page, 0, XLOG_BLCKSZ);
5230 :
5231 : /*
5232 : * Set up information for the initial checkpoint record
5233 : *
5234 : * The initial checkpoint record is written to the beginning of the WAL
5235 : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5236 : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5237 : */
5238 358 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
5239 358 : checkPoint.ThisTimeLineID = ThisTimeLineID;
5240 358 : checkPoint.PrevTimeLineID = ThisTimeLineID;
5241 358 : checkPoint.fullPageWrites = fullPageWrites;
5242 : checkPoint.nextFullXid =
5243 358 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
5244 358 : checkPoint.nextOid = FirstBootstrapObjectId;
5245 358 : checkPoint.nextMulti = FirstMultiXactId;
5246 358 : checkPoint.nextMultiOffset = 0;
5247 358 : checkPoint.oldestXid = FirstNormalTransactionId;
5248 358 : checkPoint.oldestXidDB = TemplateDbOid;
5249 358 : checkPoint.oldestMulti = FirstMultiXactId;
5250 358 : checkPoint.oldestMultiDB = TemplateDbOid;
5251 358 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5252 358 : checkPoint.newestCommitTsXid = InvalidTransactionId;
5253 358 : checkPoint.time = (pg_time_t) time(NULL);
5254 358 : checkPoint.oldestActiveXid = InvalidTransactionId;
5255 :
5256 358 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
5257 358 : ShmemVariableCache->nextOid = checkPoint.nextOid;
5258 358 : ShmemVariableCache->oidCount = 0;
5259 358 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5260 358 : AdvanceOldestClogXid(checkPoint.oldestXid);
5261 358 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5262 358 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5263 358 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5264 :
5265 : /* Set up the XLOG page header */
5266 358 : page->xlp_magic = XLOG_PAGE_MAGIC;
5267 358 : page->xlp_info = XLP_LONG_HEADER;
5268 358 : page->xlp_tli = ThisTimeLineID;
5269 358 : page->xlp_pageaddr = wal_segment_size;
5270 358 : longpage = (XLogLongPageHeader) page;
5271 358 : longpage->xlp_sysid = sysidentifier;
5272 358 : longpage->xlp_seg_size = wal_segment_size;
5273 358 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5274 :
5275 : /* Insert the initial checkpoint record */
5276 358 : recptr = ((char *) page + SizeOfXLogLongPHD);
5277 358 : record = (XLogRecord *) recptr;
5278 358 : record->xl_prev = 0;
5279 358 : record->xl_xid = InvalidTransactionId;
5280 358 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5281 358 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
5282 358 : record->xl_rmid = RM_XLOG_ID;
5283 358 : recptr += SizeOfXLogRecord;
5284 : /* fill the XLogRecordDataHeaderShort struct */
5285 358 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5286 358 : *(recptr++) = sizeof(checkPoint);
5287 358 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5288 358 : recptr += sizeof(checkPoint);
5289 : Assert(recptr - (char *) record == record->xl_tot_len);
5290 :
5291 358 : INIT_CRC32C(crc);
5292 358 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5293 358 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5294 358 : FIN_CRC32C(crc);
5295 358 : record->xl_crc = crc;
5296 :
5297 : /* Create first XLOG segment file */
5298 358 : use_existent = false;
5299 358 : openLogFile = XLogFileInit(1, &use_existent, false);
5300 :
5301 : /*
5302 : * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5303 : * close the file again in a moment.
5304 : */
5305 :
5306 : /* Write the first page with the initial record */
5307 358 : errno = 0;
5308 358 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5309 358 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5310 : {
5311 : /* if write didn't set errno, assume problem is no disk space */
5312 0 : if (errno == 0)
5313 0 : errno = ENOSPC;
5314 0 : ereport(PANIC,
5315 : (errcode_for_file_access(),
5316 : errmsg("could not write bootstrap write-ahead log file: %m")));
5317 : }
5318 358 : pgstat_report_wait_end();
5319 :
5320 358 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5321 358 : if (pg_fsync(openLogFile) != 0)
5322 0 : ereport(PANIC,
5323 : (errcode_for_file_access(),
5324 : errmsg("could not fsync bootstrap write-ahead log file: %m")));
5325 358 : pgstat_report_wait_end();
5326 :
5327 358 : if (close(openLogFile) != 0)
5328 0 : ereport(PANIC,
5329 : (errcode_for_file_access(),
5330 : errmsg("could not close bootstrap write-ahead log file: %m")));
5331 :
5332 358 : openLogFile = -1;
5333 :
5334 : /* Now create pg_control */
5335 358 : InitControlFile(sysidentifier);
5336 358 : ControlFile->time = checkPoint.time;
5337 358 : ControlFile->checkPoint = checkPoint.redo;
5338 358 : ControlFile->checkPointCopy = checkPoint;
5339 :
5340 : /* some additional ControlFile fields are set in WriteControlFile() */
5341 358 : WriteControlFile();
5342 :
5343 : /* Bootstrap the commit log, too */
5344 358 : BootStrapCLOG();
5345 358 : BootStrapCommitTs();
5346 358 : BootStrapSUBTRANS();
5347 358 : BootStrapMultiXact();
5348 :
5349 358 : pfree(buffer);
5350 :
5351 : /*
5352 : * Force control file to be read - in contrast to normal processing we'd
5353 : * otherwise never run the checks and GUC related initializations therein.
5354 : */
5355 358 : ReadControlFile();
5356 358 : }
5357 :
5358 : static char *
5359 674 : str_time(pg_time_t tnow)
5360 : {
5361 : static char buf[128];
5362 :
5363 674 : pg_strftime(buf, sizeof(buf),
5364 : "%Y-%m-%d %H:%M:%S %Z",
5365 674 : pg_localtime(&tnow, log_timezone));
5366 :
5367 674 : return buf;
5368 : }
5369 :
5370 : /*
5371 : * See if there are any recovery signal files and if so, set state for
5372 : * recovery.
5373 : *
5374 : * See if there is a recovery command file (recovery.conf), and if so
5375 : * throw an ERROR since as of PG12 we no longer recognize that.
5376 : */
5377 : static void
5378 1390 : readRecoverySignalFile(void)
5379 : {
5380 : struct stat stat_buf;
5381 :
5382 1390 : if (IsBootstrapProcessingMode())
5383 1308 : return;
5384 :
5385 : /*
5386 : * Check for old recovery API file: recovery.conf
5387 : */
5388 1032 : if (stat(RECOVERY_COMMAND_FILE, &stat_buf) == 0)
5389 0 : ereport(FATAL,
5390 : (errcode_for_file_access(),
5391 : errmsg("using recovery command file \"%s\" is not supported",
5392 : RECOVERY_COMMAND_FILE)));
5393 :
5394 : /*
5395 : * Remove unused .done file, if present. Ignore if absent.
5396 : */
5397 1032 : unlink(RECOVERY_COMMAND_DONE);
5398 :
5399 : /*
5400 : * Check for recovery signal files and if found, fsync them since they
5401 : * represent server state information. We don't sweat too much about the
5402 : * possibility of fsync failure, however.
5403 : *
5404 : * If present, standby signal file takes precedence. If neither is present
5405 : * then we won't enter archive recovery.
5406 : */
5407 1032 : if (stat(STANDBY_SIGNAL_FILE, &stat_buf) == 0)
5408 : {
5409 : int fd;
5410 :
5411 80 : fd = BasicOpenFilePerm(STANDBY_SIGNAL_FILE, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
5412 : S_IRUSR | S_IWUSR);
5413 80 : if (fd >= 0)
5414 : {
5415 80 : (void) pg_fsync(fd);
5416 80 : close(fd);
5417 : }
5418 80 : standby_signal_file_found = true;
5419 : }
5420 952 : else if (stat(RECOVERY_SIGNAL_FILE, &stat_buf) == 0)
5421 : {
5422 : int fd;
5423 :
5424 2 : fd = BasicOpenFilePerm(RECOVERY_SIGNAL_FILE, O_RDWR | PG_BINARY | get_sync_bit(sync_method),
5425 : S_IRUSR | S_IWUSR);
5426 2 : if (fd >= 0)
5427 : {
5428 2 : (void) pg_fsync(fd);
5429 2 : close(fd);
5430 : }
5431 2 : recovery_signal_file_found = true;
5432 : }
5433 :
5434 1032 : StandbyModeRequested = false;
5435 1032 : ArchiveRecoveryRequested = false;
5436 1032 : if (standby_signal_file_found)
5437 : {
5438 80 : StandbyModeRequested = true;
5439 80 : ArchiveRecoveryRequested = true;
5440 : }
5441 952 : else if (recovery_signal_file_found)
5442 : {
5443 2 : StandbyModeRequested = false;
5444 2 : ArchiveRecoveryRequested = true;
5445 : }
5446 : else
5447 950 : return;
5448 :
5449 : /*
5450 : * We don't support standby mode in standalone backends; that requires
5451 : * other processes such as the WAL receiver to be alive.
5452 : */
5453 82 : if (StandbyModeRequested && !IsUnderPostmaster)
5454 0 : ereport(FATAL,
5455 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
5456 : errmsg("standby mode is not supported by single-user servers")));
5457 : }
5458 :
5459 : static void
5460 1390 : validateRecoveryParameters(void)
5461 : {
5462 1390 : if (!ArchiveRecoveryRequested)
5463 1308 : return;
5464 :
5465 : /*
5466 : * Check for compulsory parameters
5467 : */
5468 82 : if (StandbyModeRequested)
5469 : {
5470 80 : if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
5471 4 : (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
5472 0 : ereport(WARNING,
5473 : (errmsg("specified neither primary_conninfo nor restore_command"),
5474 : errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
5475 : }
5476 : else
5477 : {
5478 2 : if (recoveryRestoreCommand == NULL ||
5479 2 : strcmp(recoveryRestoreCommand, "") == 0)
5480 0 : ereport(FATAL,
5481 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5482 : errmsg("must specify restore_command when standby mode is not enabled")));
5483 : }
5484 :
5485 : /*
5486 : * Override any inconsistent requests. Note that this is a change of
5487 : * behaviour in 9.5; prior to this we simply ignored a request to pause if
5488 : * hot_standby = off, which was surprising behaviour.
5489 : */
5490 82 : if (recoveryTargetAction == RECOVERY_TARGET_ACTION_PAUSE &&
5491 82 : !EnableHotStandby)
5492 0 : recoveryTargetAction = RECOVERY_TARGET_ACTION_SHUTDOWN;
5493 :
5494 : /*
5495 : * Final parsing of recovery_target_time string; see also
5496 : * check_recovery_target_time().
5497 : */
5498 82 : if (recoveryTarget == RECOVERY_TARGET_TIME)
5499 : {
5500 0 : recoveryTargetTime = DatumGetTimestampTz(DirectFunctionCall3(timestamptz_in,
5501 : CStringGetDatum(recovery_target_time_string),
5502 : ObjectIdGetDatum(InvalidOid),
5503 : Int32GetDatum(-1)));
5504 : }
5505 :
5506 : /*
5507 : * If user specified recovery_target_timeline, validate it or compute the
5508 : * "latest" value. We can't do this until after we've gotten the restore
5509 : * command and set InArchiveRecovery, because we need to fetch timeline
5510 : * history files from the archive.
5511 : */
5512 82 : if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_NUMERIC)
5513 : {
5514 0 : TimeLineID rtli = recoveryTargetTLIRequested;
5515 :
5516 : /* Timeline 1 does not have a history file, all else should */
5517 0 : if (rtli != 1 && !existsTimeLineHistory(rtli))
5518 0 : ereport(FATAL,
5519 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5520 : errmsg("recovery target timeline %u does not exist",
5521 : rtli)));
5522 0 : recoveryTargetTLI = rtli;
5523 : }
5524 82 : else if (recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_LATEST)
5525 : {
5526 : /* We start the "latest" search from pg_control's timeline */
5527 82 : recoveryTargetTLI = findNewestTimeLine(recoveryTargetTLI);
5528 : }
5529 : else
5530 : {
5531 : /*
5532 : * else we just use the recoveryTargetTLI as already read from
5533 : * ControlFile
5534 : */
5535 : Assert(recoveryTargetTimeLineGoal == RECOVERY_TARGET_TIMELINE_CONTROLFILE);
5536 : }
5537 : }
5538 :
5539 : /*
5540 : * Exit archive-recovery state
5541 : */
5542 : static void
5543 48 : exitArchiveRecovery(TimeLineID endTLI, XLogRecPtr endOfLog)
5544 : {
5545 : char xlogfname[MAXFNAMELEN];
5546 : XLogSegNo endLogSegNo;
5547 : XLogSegNo startLogSegNo;
5548 :
5549 : /* we always switch to a new timeline after archive recovery */
5550 : Assert(endTLI != ThisTimeLineID);
5551 :
5552 : /*
5553 : * We are no longer in archive recovery state.
5554 : */
5555 48 : InArchiveRecovery = false;
5556 :
5557 : /*
5558 : * Update min recovery point one last time.
5559 : */
5560 48 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5561 :
5562 : /*
5563 : * If the ending log segment is still open, close it (to avoid problems on
5564 : * Windows with trying to rename or delete an open file).
5565 : */
5566 48 : if (readFile >= 0)
5567 : {
5568 48 : close(readFile);
5569 48 : readFile = -1;
5570 : }
5571 :
5572 : /*
5573 : * Calculate the last segment on the old timeline, and the first segment
5574 : * on the new timeline. If the switch happens in the middle of a segment,
5575 : * they are the same, but if the switch happens exactly at a segment
5576 : * boundary, startLogSegNo will be endLogSegNo + 1.
5577 : */
5578 48 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5579 48 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5580 :
5581 : /*
5582 : * Initialize the starting WAL segment for the new timeline. If the switch
5583 : * happens in the middle of a segment, copy data from the last WAL segment
5584 : * of the old timeline up to the switch point, to the starting WAL segment
5585 : * on the new timeline.
5586 : */
5587 48 : if (endLogSegNo == startLogSegNo)
5588 : {
5589 : /*
5590 : * Make a copy of the file on the new timeline.
5591 : *
5592 : * Writing WAL isn't allowed yet, so there are no locking
5593 : * considerations. But we should be just as tense as XLogFileInit to
5594 : * avoid emplacing a bogus file.
5595 : */
5596 36 : XLogFileCopy(endLogSegNo, endTLI, endLogSegNo,
5597 36 : XLogSegmentOffset(endOfLog, wal_segment_size));
5598 : }
5599 : else
5600 : {
5601 : /*
5602 : * The switch happened at a segment boundary, so just create the next
5603 : * segment on the new timeline.
5604 : */
5605 12 : bool use_existent = true;
5606 : int fd;
5607 :
5608 12 : fd = XLogFileInit(startLogSegNo, &use_existent, true);
5609 :
5610 12 : if (close(fd) != 0)
5611 : {
5612 : char xlogfname[MAXFNAMELEN];
5613 0 : int save_errno = errno;
5614 :
5615 0 : XLogFileName(xlogfname, ThisTimeLineID, startLogSegNo,
5616 : wal_segment_size);
5617 0 : errno = save_errno;
5618 0 : ereport(ERROR,
5619 : (errcode_for_file_access(),
5620 : errmsg("could not close file \"%s\": %m", xlogfname)));
5621 : }
5622 : }
5623 :
5624 : /*
5625 : * Let's just make real sure there are not .ready or .done flags posted
5626 : * for the new segment.
5627 : */
5628 48 : XLogFileName(xlogfname, ThisTimeLineID, startLogSegNo, wal_segment_size);
5629 48 : XLogArchiveCleanup(xlogfname);
5630 :
5631 : /*
5632 : * Remove the signal files out of the way, so that we don't accidentally
5633 : * re-enter archive recovery mode in a subsequent crash.
5634 : */
5635 48 : if (standby_signal_file_found)
5636 48 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
5637 :
5638 48 : if (recovery_signal_file_found)
5639 0 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
5640 :
5641 48 : ereport(LOG,
5642 : (errmsg("archive recovery complete")));
5643 48 : }
5644 :
5645 : /*
5646 : * Extract timestamp from WAL record.
5647 : *
5648 : * If the record contains a timestamp, returns true, and saves the timestamp
5649 : * in *recordXtime. If the record type has no timestamp, returns false.
5650 : * Currently, only transaction commit/abort records and restore points contain
5651 : * timestamps.
5652 : */
5653 : static bool
5654 264 : getRecordTimestamp(XLogReaderState *record, TimestampTz *recordXtime)
5655 : {
5656 264 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
5657 264 : uint8 xact_info = info & XLOG_XACT_OPMASK;
5658 264 : uint8 rmid = XLogRecGetRmid(record);
5659 :
5660 264 : if (rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
5661 : {
5662 0 : *recordXtime = ((xl_restore_point *) XLogRecGetData(record))->rp_time;
5663 0 : return true;
5664 : }
5665 264 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_COMMIT ||
5666 : xact_info == XLOG_XACT_COMMIT_PREPARED))
5667 : {
5668 256 : *recordXtime = ((xl_xact_commit *) XLogRecGetData(record))->xact_time;
5669 256 : return true;
5670 : }
5671 8 : if (rmid == RM_XACT_ID && (xact_info == XLOG_XACT_ABORT ||
5672 : xact_info == XLOG_XACT_ABORT_PREPARED))
5673 : {
5674 8 : *recordXtime = ((xl_xact_abort *) XLogRecGetData(record))->xact_time;
5675 8 : return true;
5676 : }
5677 0 : return false;
5678 : }
5679 :
5680 : /*
5681 : * For point-in-time recovery, this function decides whether we want to
5682 : * stop applying the XLOG before the current record.
5683 : *
5684 : * Returns true if we are stopping, false otherwise. If stopping, some
5685 : * information is saved in recoveryStopXid et al for use in annotating the
5686 : * new timeline's history file.
5687 : */
5688 : static bool
5689 573454 : recoveryStopsBefore(XLogReaderState *record)
5690 : {
5691 573454 : bool stopsHere = false;
5692 : uint8 xact_info;
5693 : bool isCommit;
5694 573454 : TimestampTz recordXtime = 0;
5695 : TransactionId recordXid;
5696 :
5697 : /*
5698 : * Ignore recovery target settings when not in archive recovery (meaning
5699 : * we are in crash recovery).
5700 : */
5701 573454 : if (!ArchiveRecoveryRequested)
5702 370862 : return false;
5703 :
5704 : /* Check if we should stop as soon as reaching consistency */
5705 202592 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
5706 : {
5707 0 : ereport(LOG,
5708 : (errmsg("recovery stopping after reaching consistency")));
5709 :
5710 0 : recoveryStopAfter = false;
5711 0 : recoveryStopXid = InvalidTransactionId;
5712 0 : recoveryStopLSN = InvalidXLogRecPtr;
5713 0 : recoveryStopTime = 0;
5714 0 : recoveryStopName[0] = '\0';
5715 0 : return true;
5716 : }
5717 :
5718 : /* Check if target LSN has been reached */
5719 202592 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
5720 0 : !recoveryTargetInclusive &&
5721 0 : record->ReadRecPtr >= recoveryTargetLSN)
5722 : {
5723 0 : recoveryStopAfter = false;
5724 0 : recoveryStopXid = InvalidTransactionId;
5725 0 : recoveryStopLSN = record->ReadRecPtr;
5726 0 : recoveryStopTime = 0;
5727 0 : recoveryStopName[0] = '\0';
5728 0 : ereport(LOG,
5729 : (errmsg("recovery stopping before WAL location (LSN) \"%X/%X\"",
5730 : (uint32) (recoveryStopLSN >> 32),
5731 : (uint32) recoveryStopLSN)));
5732 0 : return true;
5733 : }
5734 :
5735 : /* Otherwise we only consider stopping before COMMIT or ABORT records. */
5736 202592 : if (XLogRecGetRmid(record) != RM_XACT_ID)
5737 202246 : return false;
5738 :
5739 346 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
5740 :
5741 346 : if (xact_info == XLOG_XACT_COMMIT)
5742 : {
5743 226 : isCommit = true;
5744 226 : recordXid = XLogRecGetXid(record);
5745 : }
5746 120 : else if (xact_info == XLOG_XACT_COMMIT_PREPARED)
5747 : {
5748 30 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
5749 : xl_xact_parsed_commit parsed;
5750 :
5751 30 : isCommit = true;
5752 30 : ParseCommitRecord(XLogRecGetInfo(record),
5753 : xlrec,
5754 : &parsed);
5755 30 : recordXid = parsed.twophase_xid;
5756 : }
5757 90 : else if (xact_info == XLOG_XACT_ABORT)
5758 : {
5759 2 : isCommit = false;
5760 2 : recordXid = XLogRecGetXid(record);
5761 : }
5762 88 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
5763 : {
5764 6 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
5765 : xl_xact_parsed_abort parsed;
5766 :
5767 6 : isCommit = true;
5768 6 : ParseAbortRecord(XLogRecGetInfo(record),
5769 : xlrec,
5770 : &parsed);
5771 6 : recordXid = parsed.twophase_xid;
5772 : }
5773 : else
5774 82 : return false;
5775 :
5776 264 : if (recoveryTarget == RECOVERY_TARGET_XID && !recoveryTargetInclusive)
5777 : {
5778 : /*
5779 : * There can be only one transaction end record with this exact
5780 : * transactionid
5781 : *
5782 : * when testing for an xid, we MUST test for equality only, since
5783 : * transactions are numbered in the order they start, not the order
5784 : * they complete. A higher numbered xid will complete before you about
5785 : * 50% of the time...
5786 : */
5787 0 : stopsHere = (recordXid == recoveryTargetXid);
5788 : }
5789 :
5790 264 : if (recoveryTarget == RECOVERY_TARGET_TIME &&
5791 0 : getRecordTimestamp(record, &recordXtime))
5792 : {
5793 : /*
5794 : * There can be many transactions that share the same commit time, so
5795 : * we stop after the last one, if we are inclusive, or stop at the
5796 : * first one if we are exclusive
5797 : */
5798 0 : if (recoveryTargetInclusive)
5799 0 : stopsHere = (recordXtime > recoveryTargetTime);
5800 : else
5801 0 : stopsHere = (recordXtime >= recoveryTargetTime);
5802 : }
5803 :
5804 264 : if (stopsHere)
5805 : {
5806 0 : recoveryStopAfter = false;
5807 0 : recoveryStopXid = recordXid;
5808 0 : recoveryStopTime = recordXtime;
5809 0 : recoveryStopLSN = InvalidXLogRecPtr;
5810 0 : recoveryStopName[0] = '\0';
5811 :
5812 0 : if (isCommit)
5813 : {
5814 0 : ereport(LOG,
5815 : (errmsg("recovery stopping before commit of transaction %u, time %s",
5816 : recoveryStopXid,
5817 : timestamptz_to_str(recoveryStopTime))));
5818 : }
5819 : else
5820 : {
5821 0 : ereport(LOG,
5822 : (errmsg("recovery stopping before abort of transaction %u, time %s",
5823 : recoveryStopXid,
5824 : timestamptz_to_str(recoveryStopTime))));
5825 : }
5826 : }
5827 :
5828 264 : return stopsHere;
5829 : }
5830 :
5831 : /*
5832 : * Same as recoveryStopsBefore, but called after applying the record.
5833 : *
5834 : * We also track the timestamp of the latest applied COMMIT/ABORT
5835 : * record in XLogCtl->recoveryLastXTime.
5836 : */
5837 : static bool
5838 573454 : recoveryStopsAfter(XLogReaderState *record)
5839 : {
5840 : uint8 info;
5841 : uint8 xact_info;
5842 : uint8 rmid;
5843 : TimestampTz recordXtime;
5844 :
5845 : /*
5846 : * Ignore recovery target settings when not in archive recovery (meaning
5847 : * we are in crash recovery).
5848 : */
5849 573454 : if (!ArchiveRecoveryRequested)
5850 370862 : return false;
5851 :
5852 202592 : info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
5853 202592 : rmid = XLogRecGetRmid(record);
5854 :
5855 : /*
5856 : * There can be many restore points that share the same name; we stop at
5857 : * the first one.
5858 : */
5859 202592 : if (recoveryTarget == RECOVERY_TARGET_NAME &&
5860 14 : rmid == RM_XLOG_ID && info == XLOG_RESTORE_POINT)
5861 : {
5862 : xl_restore_point *recordRestorePointData;
5863 :
5864 2 : recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
5865 :
5866 2 : if (strcmp(recordRestorePointData->rp_name, recoveryTargetName) == 0)
5867 : {
5868 0 : recoveryStopAfter = true;
5869 0 : recoveryStopXid = InvalidTransactionId;
5870 0 : recoveryStopLSN = InvalidXLogRecPtr;
5871 0 : (void) getRecordTimestamp(record, &recoveryStopTime);
5872 0 : strlcpy(recoveryStopName, recordRestorePointData->rp_name, MAXFNAMELEN);
5873 :
5874 0 : ereport(LOG,
5875 : (errmsg("recovery stopping at restore point \"%s\", time %s",
5876 : recoveryStopName,
5877 : timestamptz_to_str(recoveryStopTime))));
5878 0 : return true;
5879 : }
5880 : }
5881 :
5882 : /* Check if the target LSN has been reached */
5883 202592 : if (recoveryTarget == RECOVERY_TARGET_LSN &&
5884 0 : recoveryTargetInclusive &&
5885 0 : record->ReadRecPtr >= recoveryTargetLSN)
5886 : {
5887 0 : recoveryStopAfter = true;
5888 0 : recoveryStopXid = InvalidTransactionId;
5889 0 : recoveryStopLSN = record->ReadRecPtr;
5890 0 : recoveryStopTime = 0;
5891 0 : recoveryStopName[0] = '\0';
5892 0 : ereport(LOG,
5893 : (errmsg("recovery stopping after WAL location (LSN) \"%X/%X\"",
5894 : (uint32) (recoveryStopLSN >> 32),
5895 : (uint32) recoveryStopLSN)));
5896 0 : return true;
5897 : }
5898 :
5899 202592 : if (rmid != RM_XACT_ID)
5900 202246 : return false;
5901 :
5902 346 : xact_info = info & XLOG_XACT_OPMASK;
5903 :
5904 346 : if (xact_info == XLOG_XACT_COMMIT ||
5905 90 : xact_info == XLOG_XACT_COMMIT_PREPARED ||
5906 88 : xact_info == XLOG_XACT_ABORT ||
5907 : xact_info == XLOG_XACT_ABORT_PREPARED)
5908 : {
5909 : TransactionId recordXid;
5910 :
5911 : /* Update the last applied transaction timestamp */
5912 264 : if (getRecordTimestamp(record, &recordXtime))
5913 264 : SetLatestXTime(recordXtime);
5914 :
5915 : /* Extract the XID of the committed/aborted transaction */
5916 264 : if (xact_info == XLOG_XACT_COMMIT_PREPARED)
5917 : {
5918 30 : xl_xact_commit *xlrec = (xl_xact_commit *) XLogRecGetData(record);
5919 : xl_xact_parsed_commit parsed;
5920 :
5921 30 : ParseCommitRecord(XLogRecGetInfo(record),
5922 : xlrec,
5923 : &parsed);
5924 30 : recordXid = parsed.twophase_xid;
5925 : }
5926 234 : else if (xact_info == XLOG_XACT_ABORT_PREPARED)
5927 : {
5928 6 : xl_xact_abort *xlrec = (xl_xact_abort *) XLogRecGetData(record);
5929 : xl_xact_parsed_abort parsed;
5930 :
5931 6 : ParseAbortRecord(XLogRecGetInfo(record),
5932 : xlrec,
5933 : &parsed);
5934 6 : recordXid = parsed.twophase_xid;
5935 : }
5936 : else
5937 228 : recordXid = XLogRecGetXid(record);
5938 :
5939 : /*
5940 : * There can be only one transaction end record with this exact
5941 : * transactionid
5942 : *
5943 : * when testing for an xid, we MUST test for equality only, since
5944 : * transactions are numbered in the order they start, not the order
5945 : * they complete. A higher numbered xid will complete before you about
5946 : * 50% of the time...
5947 : */
5948 264 : if (recoveryTarget == RECOVERY_TARGET_XID && recoveryTargetInclusive &&
5949 0 : recordXid == recoveryTargetXid)
5950 : {
5951 0 : recoveryStopAfter = true;
5952 0 : recoveryStopXid = recordXid;
5953 0 : recoveryStopTime = recordXtime;
5954 0 : recoveryStopLSN = InvalidXLogRecPtr;
5955 0 : recoveryStopName[0] = '\0';
5956 :
5957 0 : if (xact_info == XLOG_XACT_COMMIT ||
5958 : xact_info == XLOG_XACT_COMMIT_PREPARED)
5959 : {
5960 0 : ereport(LOG,
5961 : (errmsg("recovery stopping after commit of transaction %u, time %s",
5962 : recoveryStopXid,
5963 : timestamptz_to_str(recoveryStopTime))));
5964 : }
5965 0 : else if (xact_info == XLOG_XACT_ABORT ||
5966 : xact_info == XLOG_XACT_ABORT_PREPARED)
5967 : {
5968 0 : ereport(LOG,
5969 : (errmsg("recovery stopping after abort of transaction %u, time %s",
5970 : recoveryStopXid,
5971 : timestamptz_to_str(recoveryStopTime))));
5972 : }
5973 0 : return true;
5974 : }
5975 : }
5976 :
5977 : /* Check if we should stop as soon as reaching consistency */
5978 346 : if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE && reachedConsistency)
5979 : {
5980 0 : ereport(LOG,
5981 : (errmsg("recovery stopping after reaching consistency")));
5982 :
5983 0 : recoveryStopAfter = true;
5984 0 : recoveryStopXid = InvalidTransactionId;
5985 0 : recoveryStopTime = 0;
5986 0 : recoveryStopLSN = InvalidXLogRecPtr;
5987 0 : recoveryStopName[0] = '\0';
5988 0 : return true;
5989 : }
5990 :
5991 346 : return false;
5992 : }
5993 :
5994 : /*
5995 : * Wait until shared recoveryPause flag is cleared.
5996 : *
5997 : * endOfRecovery is true if the recovery target is reached and
5998 : * the paused state starts at the end of recovery because of
5999 : * recovery_target_action=pause, and false otherwise.
6000 : *
6001 : * XXX Could also be done with shared latch, avoiding the pg_usleep loop.
6002 : * Probably not worth the trouble though. This state shouldn't be one that
6003 : * anyone cares about server power consumption in.
6004 : */
6005 : static void
6006 0 : recoveryPausesHere(bool endOfRecovery)
6007 : {
6008 : /* Don't pause unless users can connect! */
6009 0 : if (!LocalHotStandbyActive)
6010 0 : return;
6011 :
6012 : /* Don't pause after standby promotion has been triggered */
6013 0 : if (LocalPromoteIsTriggered)
6014 0 : return;
6015 :
6016 0 : if (endOfRecovery)
6017 0 : ereport(LOG,
6018 : (errmsg("pausing at the end of recovery"),
6019 : errhint("Execute pg_wal_replay_resume() to promote.")));
6020 : else
6021 0 : ereport(LOG,
6022 : (errmsg("recovery has paused"),
6023 : errhint("Execute pg_wal_replay_resume() to continue.")));
6024 :
6025 0 : while (RecoveryIsPaused())
6026 : {
6027 0 : HandleStartupProcInterrupts();
6028 0 : if (CheckForStandbyTrigger())
6029 0 : return;
6030 0 : pgstat_report_wait_start(WAIT_EVENT_RECOVERY_PAUSE);
6031 0 : pg_usleep(1000000L); /* 1000 ms */
6032 0 : pgstat_report_wait_end();
6033 : }
6034 : }
6035 :
6036 : bool
6037 0 : RecoveryIsPaused(void)
6038 : {
6039 : bool recoveryPause;
6040 :
6041 0 : SpinLockAcquire(&XLogCtl->info_lck);
6042 0 : recoveryPause = XLogCtl->recoveryPause;
6043 0 : SpinLockRelease(&XLogCtl->info_lck);
6044 :
6045 0 : return recoveryPause;
6046 : }
6047 :
6048 : void
6049 0 : SetRecoveryPause(bool recoveryPause)
6050 : {
6051 0 : SpinLockAcquire(&XLogCtl->info_lck);
6052 0 : XLogCtl->recoveryPause = recoveryPause;
6053 0 : SpinLockRelease(&XLogCtl->info_lck);
6054 0 : }
6055 :
6056 : /*
6057 : * When recovery_min_apply_delay is set, we wait long enough to make sure
6058 : * certain record types are applied at least that interval behind the master.
6059 : *
6060 : * Returns true if we waited.
6061 : *
6062 : * Note that the delay is calculated between the WAL record log time and
6063 : * the current time on standby. We would prefer to keep track of when this
6064 : * standby received each WAL record, which would allow a more consistent
6065 : * approach and one not affected by time synchronisation issues, but that
6066 : * is significantly more effort and complexity for little actual gain in
6067 : * usability.
6068 : */
6069 : static bool
6070 573454 : recoveryApplyDelay(XLogReaderState *record)
6071 : {
6072 : uint8 xact_info;
6073 : TimestampTz xtime;
6074 : TimestampTz delayUntil;
6075 : long secs;
6076 : int microsecs;
6077 :
6078 : /* nothing to do if no delay configured */
6079 573454 : if (recovery_min_apply_delay <= 0)
6080 573454 : return false;
6081 :
6082 : /* no delay is applied on a database not yet consistent */
6083 0 : if (!reachedConsistency)
6084 0 : return false;
6085 :
6086 : /* nothing to do if crash recovery is requested */
6087 0 : if (!ArchiveRecoveryRequested)
6088 0 : return false;
6089 :
6090 : /*
6091 : * Is it a COMMIT record?
6092 : *
6093 : * We deliberately choose not to delay aborts since they have no effect on
6094 : * MVCC. We already allow replay of records that don't have a timestamp,
6095 : * so there is already opportunity for issues caused by early conflicts on
6096 : * standbys.
6097 : */
6098 0 : if (XLogRecGetRmid(record) != RM_XACT_ID)
6099 0 : return false;
6100 :
6101 0 : xact_info = XLogRecGetInfo(record) & XLOG_XACT_OPMASK;
6102 :
6103 0 : if (xact_info != XLOG_XACT_COMMIT &&
6104 : xact_info != XLOG_XACT_COMMIT_PREPARED)
6105 0 : return false;
6106 :
6107 0 : if (!getRecordTimestamp(record, &xtime))
6108 0 : return false;
6109 :
6110 0 : delayUntil = TimestampTzPlusMilliseconds(xtime, recovery_min_apply_delay);
6111 :
6112 : /*
6113 : * Exit without arming the latch if it's already past time to apply this
6114 : * record
6115 : */
6116 0 : TimestampDifference(GetCurrentTimestamp(), delayUntil, &secs, µsecs);
6117 0 : if (secs <= 0 && microsecs <= 0)
6118 0 : return false;
6119 :
6120 : while (true)
6121 : {
6122 0 : ResetLatch(&XLogCtl->recoveryWakeupLatch);
6123 :
6124 : /* might change the trigger file's location */
6125 0 : HandleStartupProcInterrupts();
6126 :
6127 0 : if (CheckForStandbyTrigger())
6128 0 : break;
6129 :
6130 : /*
6131 : * Wait for difference between GetCurrentTimestamp() and delayUntil
6132 : */
6133 0 : TimestampDifference(GetCurrentTimestamp(), delayUntil,
6134 : &secs, µsecs);
6135 :
6136 : /*
6137 : * NB: We're ignoring waits below recovery_min_apply_delay's
6138 : * resolution.
6139 : */
6140 0 : if (secs <= 0 && microsecs / 1000 <= 0)
6141 0 : break;
6142 :
6143 0 : elog(DEBUG2, "recovery apply delay %ld seconds, %d milliseconds",
6144 : secs, microsecs / 1000);
6145 :
6146 0 : (void) WaitLatch(&XLogCtl->recoveryWakeupLatch,
6147 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
6148 0 : secs * 1000L + microsecs / 1000,
6149 : WAIT_EVENT_RECOVERY_APPLY_DELAY);
6150 : }
6151 0 : return true;
6152 : }
6153 :
6154 : /*
6155 : * Save timestamp of latest processed commit/abort record.
6156 : *
6157 : * We keep this in XLogCtl, not a simple static variable, so that it can be
6158 : * seen by processes other than the startup process. Note in particular
6159 : * that CreateRestartPoint is executed in the checkpointer.
6160 : */
6161 : static void
6162 264 : SetLatestXTime(TimestampTz xtime)
6163 : {
6164 264 : SpinLockAcquire(&XLogCtl->info_lck);
6165 264 : XLogCtl->recoveryLastXTime = xtime;
6166 264 : SpinLockRelease(&XLogCtl->info_lck);
6167 264 : }
6168 :
6169 : /*
6170 : * Fetch timestamp of latest processed commit/abort record.
6171 : */
6172 : TimestampTz
6173 186 : GetLatestXTime(void)
6174 : {
6175 : TimestampTz xtime;
6176 :
6177 186 : SpinLockAcquire(&XLogCtl->info_lck);
6178 186 : xtime = XLogCtl->recoveryLastXTime;
6179 186 : SpinLockRelease(&XLogCtl->info_lck);
6180 :
6181 186 : return xtime;
6182 : }
6183 :
6184 : /*
6185 : * Save timestamp of the next chunk of WAL records to apply.
6186 : *
6187 : * We keep this in XLogCtl, not a simple static variable, so that it can be
6188 : * seen by all backends.
6189 : */
6190 : static void
6191 394 : SetCurrentChunkStartTime(TimestampTz xtime)
6192 : {
6193 394 : SpinLockAcquire(&XLogCtl->info_lck);
6194 394 : XLogCtl->currentChunkStartTime = xtime;
6195 394 : SpinLockRelease(&XLogCtl->info_lck);
6196 394 : }
6197 :
6198 : /*
6199 : * Fetch timestamp of latest processed commit/abort record.
6200 : * Startup process maintains an accurate local copy in XLogReceiptTime
6201 : */
6202 : TimestampTz
6203 0 : GetCurrentChunkReplayStartTime(void)
6204 : {
6205 : TimestampTz xtime;
6206 :
6207 0 : SpinLockAcquire(&XLogCtl->info_lck);
6208 0 : xtime = XLogCtl->currentChunkStartTime;
6209 0 : SpinLockRelease(&XLogCtl->info_lck);
6210 :
6211 0 : return xtime;
6212 : }
6213 :
6214 : /*
6215 : * Returns time of receipt of current chunk of XLOG data, as well as
6216 : * whether it was received from streaming replication or from archives.
6217 : */
6218 : void
6219 0 : GetXLogReceiptTime(TimestampTz *rtime, bool *fromStream)
6220 : {
6221 : /*
6222 : * This must be executed in the startup process, since we don't export the
6223 : * relevant state to shared memory.
6224 : */
6225 : Assert(InRecovery);
6226 :
6227 0 : *rtime = XLogReceiptTime;
6228 0 : *fromStream = (XLogReceiptSource == XLOG_FROM_STREAM);
6229 0 : }
6230 :
6231 : /*
6232 : * Note that text field supplied is a parameter name and does not require
6233 : * translation
6234 : */
6235 : #define RecoveryRequiresIntParameter(param_name, currValue, minValue) \
6236 : do { \
6237 : if ((currValue) < (minValue)) \
6238 : ereport(ERROR, \
6239 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE), \
6240 : errmsg("hot standby is not possible because %s = %d is a lower setting than on the master server (its value was %d)", \
6241 : param_name, \
6242 : currValue, \
6243 : minValue))); \
6244 : } while(0)
6245 :
6246 : /*
6247 : * Check to see if required parameters are set high enough on this server
6248 : * for various aspects of recovery operation.
6249 : *
6250 : * Note that all the parameters which this function tests need to be
6251 : * listed in Administrator's Overview section in high-availability.sgml.
6252 : * If you change them, don't forget to update the list.
6253 : */
6254 : static void
6255 214 : CheckRequiredParameterValues(void)
6256 : {
6257 : /*
6258 : * For archive recovery, the WAL must be generated with at least 'replica'
6259 : * wal_level.
6260 : */
6261 214 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
6262 : {
6263 0 : ereport(WARNING,
6264 : (errmsg("WAL was generated with wal_level=minimal, data may be missing"),
6265 : errhint("This happens if you temporarily set wal_level=minimal without taking a new base backup.")));
6266 : }
6267 :
6268 : /*
6269 : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
6270 : * must have at least as many backend slots as the primary.
6271 : */
6272 214 : if (ArchiveRecoveryRequested && EnableHotStandby)
6273 : {
6274 86 : if (ControlFile->wal_level < WAL_LEVEL_REPLICA)
6275 0 : ereport(ERROR,
6276 : (errmsg("hot standby is not possible because wal_level was not set to \"replica\" or higher on the master server"),
6277 : errhint("Either set wal_level to \"replica\" on the master, or turn off hot_standby here.")));
6278 :
6279 : /* We ignore autovacuum_max_workers when we make this test. */
6280 86 : RecoveryRequiresIntParameter("max_connections",
6281 : MaxConnections,
6282 : ControlFile->MaxConnections);
6283 86 : RecoveryRequiresIntParameter("max_worker_processes",
6284 : max_worker_processes,
6285 : ControlFile->max_worker_processes);
6286 86 : RecoveryRequiresIntParameter("max_wal_senders",
6287 : max_wal_senders,
6288 : ControlFile->max_wal_senders);
6289 86 : RecoveryRequiresIntParameter("max_prepared_transactions",
6290 : max_prepared_xacts,
6291 : ControlFile->max_prepared_xacts);
6292 86 : RecoveryRequiresIntParameter("max_locks_per_transaction",
6293 : max_locks_per_xact,
6294 : ControlFile->max_locks_per_xact);
6295 : }
6296 214 : }
6297 :
6298 : /*
6299 : * This must be called ONCE during postmaster or standalone-backend startup
6300 : */
6301 : void
6302 1390 : StartupXLOG(void)
6303 : {
6304 : XLogCtlInsert *Insert;
6305 : CheckPoint checkPoint;
6306 : bool wasShutdown;
6307 1390 : bool reachedRecoveryTarget = false;
6308 1390 : bool haveBackupLabel = false;
6309 1390 : bool haveTblspcMap = false;
6310 : XLogRecPtr RecPtr,
6311 : checkPointLoc,
6312 : EndOfLog;
6313 : TimeLineID EndOfLogTLI;
6314 : TimeLineID PrevTimeLineID;
6315 : XLogRecord *record;
6316 : TransactionId oldestActiveXID;
6317 1390 : bool backupEndRequired = false;
6318 1390 : bool backupFromStandby = false;
6319 : DBState dbstate_at_startup;
6320 : XLogReaderState *xlogreader;
6321 : XLogPageReadPrivate private;
6322 1390 : bool fast_promoted = false;
6323 : struct stat st;
6324 :
6325 : /*
6326 : * We should have an aux process resource owner to use, and we should not
6327 : * be in a transaction that's installed some other resowner.
6328 : */
6329 : Assert(AuxProcessResourceOwner != NULL);
6330 : Assert(CurrentResourceOwner == NULL ||
6331 : CurrentResourceOwner == AuxProcessResourceOwner);
6332 1390 : CurrentResourceOwner = AuxProcessResourceOwner;
6333 :
6334 : /*
6335 : * Check that contents look valid.
6336 : */
6337 1390 : if (!XRecOffIsValid(ControlFile->checkPoint))
6338 0 : ereport(FATAL,
6339 : (errmsg("control file contains invalid checkpoint location")));
6340 :
6341 1390 : switch (ControlFile->state)
6342 : {
6343 1198 : case DB_SHUTDOWNED:
6344 :
6345 : /*
6346 : * This is the expected case, so don't be chatty in standalone
6347 : * mode
6348 : */
6349 1198 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6350 : (errmsg("database system was shut down at %s",
6351 : str_time(ControlFile->time))));
6352 1198 : break;
6353 :
6354 16 : case DB_SHUTDOWNED_IN_RECOVERY:
6355 16 : ereport(LOG,
6356 : (errmsg("database system was shut down in recovery at %s",
6357 : str_time(ControlFile->time))));
6358 16 : break;
6359 :
6360 0 : case DB_SHUTDOWNING:
6361 0 : ereport(LOG,
6362 : (errmsg("database system shutdown was interrupted; last known up at %s",
6363 : str_time(ControlFile->time))));
6364 0 : break;
6365 :
6366 0 : case DB_IN_CRASH_RECOVERY:
6367 0 : ereport(LOG,
6368 : (errmsg("database system was interrupted while in recovery at %s",
6369 : str_time(ControlFile->time)),
6370 : errhint("This probably means that some data is corrupted and"
6371 : " you will have to use the last backup for recovery.")));
6372 0 : break;
6373 :
6374 2 : case DB_IN_ARCHIVE_RECOVERY:
6375 2 : ereport(LOG,
6376 : (errmsg("database system was interrupted while in recovery at log time %s",
6377 : str_time(ControlFile->checkPointCopy.time)),
6378 : errhint("If this has occurred more than once some data might be corrupted"
6379 : " and you might need to choose an earlier recovery target.")));
6380 2 : break;
6381 :
6382 174 : case DB_IN_PRODUCTION:
6383 174 : ereport(LOG,
6384 : (errmsg("database system was interrupted; last known up at %s",
6385 : str_time(ControlFile->time))));
6386 174 : break;
6387 :
6388 0 : default:
6389 0 : ereport(FATAL,
6390 : (errmsg("control file contains invalid database cluster state")));
6391 : }
6392 :
6393 : /* This is just to allow attaching to startup process with a debugger */
6394 : #ifdef XLOG_REPLAY_DELAY
6395 : if (ControlFile->state != DB_SHUTDOWNED)
6396 : pg_usleep(60000000L);
6397 : #endif
6398 :
6399 : /*
6400 : * Verify that pg_wal and pg_wal/archive_status exist. In cases where
6401 : * someone has performed a copy for PITR, these directories may have been
6402 : * excluded and need to be re-created.
6403 : */
6404 1390 : ValidateXLOGDirectoryStructure();
6405 :
6406 : /*----------
6407 : * If we previously crashed, perform a couple of actions:
6408 : *
6409 : * - The pg_wal directory may still include some temporary WAL segments
6410 : * used when creating a new segment, so perform some clean up to not
6411 : * bloat this path. This is done first as there is no point to sync
6412 : * this temporary data.
6413 : *
6414 : * - There might be data which we had written, intending to fsync it, but
6415 : * which we had not actually fsync'd yet. Therefore, a power failure in
6416 : * the near future might cause earlier unflushed writes to be lost, even
6417 : * though more recent data written to disk from here on would be
6418 : * persisted. To avoid that, fsync the entire data directory.
6419 : */
6420 1390 : if (ControlFile->state != DB_SHUTDOWNED &&
6421 192 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
6422 : {
6423 176 : RemoveTempXlogFiles();
6424 176 : SyncDataDirectory();
6425 : }
6426 :
6427 : /*
6428 : * Initialize on the assumption we want to recover to the latest timeline
6429 : * that's active according to pg_control.
6430 : */
6431 2780 : if (ControlFile->minRecoveryPointTLI >
6432 1390 : ControlFile->checkPointCopy.ThisTimeLineID)
6433 2 : recoveryTargetTLI = ControlFile->minRecoveryPointTLI;
6434 : else
6435 1388 : recoveryTargetTLI = ControlFile->checkPointCopy.ThisTimeLineID;
6436 :
6437 : /*
6438 : * Check for signal files, and if so set up state for offline recovery
6439 : */
6440 1390 : readRecoverySignalFile();
6441 1390 : validateRecoveryParameters();
6442 :
6443 1390 : if (ArchiveRecoveryRequested)
6444 : {
6445 82 : if (StandbyModeRequested)
6446 80 : ereport(LOG,
6447 : (errmsg("entering standby mode")));
6448 2 : else if (recoveryTarget == RECOVERY_TARGET_XID)
6449 0 : ereport(LOG,
6450 : (errmsg("starting point-in-time recovery to XID %u",
6451 : recoveryTargetXid)));
6452 2 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
6453 0 : ereport(LOG,
6454 : (errmsg("starting point-in-time recovery to %s",
6455 : timestamptz_to_str(recoveryTargetTime))));
6456 2 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
6457 2 : ereport(LOG,
6458 : (errmsg("starting point-in-time recovery to \"%s\"",
6459 : recoveryTargetName)));
6460 0 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
6461 0 : ereport(LOG,
6462 : (errmsg("starting point-in-time recovery to WAL location (LSN) \"%X/%X\"",
6463 : (uint32) (recoveryTargetLSN >> 32),
6464 : (uint32) recoveryTargetLSN)));
6465 0 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
6466 0 : ereport(LOG,
6467 : (errmsg("starting point-in-time recovery to earliest consistent point")));
6468 : else
6469 0 : ereport(LOG,
6470 : (errmsg("starting archive recovery")));
6471 : }
6472 :
6473 : /*
6474 : * Take ownership of the wakeup latch if we're going to sleep during
6475 : * recovery.
6476 : */
6477 1390 : if (ArchiveRecoveryRequested)
6478 82 : OwnLatch(&XLogCtl->recoveryWakeupLatch);
6479 :
6480 : /* Set up XLOG reader facility */
6481 1390 : MemSet(&private, 0, sizeof(XLogPageReadPrivate));
6482 : xlogreader =
6483 1390 : XLogReaderAllocate(wal_segment_size, NULL,
6484 1390 : XL_ROUTINE(.page_read = &XLogPageRead,
6485 : .segment_open = NULL,
6486 : .segment_close = wal_segment_close),
6487 : &private);
6488 1390 : if (!xlogreader)
6489 0 : ereport(ERROR,
6490 : (errcode(ERRCODE_OUT_OF_MEMORY),
6491 : errmsg("out of memory"),
6492 : errdetail("Failed while allocating a WAL reading processor.")));
6493 1390 : xlogreader->system_identifier = ControlFile->system_identifier;
6494 :
6495 : /*
6496 : * Allocate two page buffers dedicated to WAL consistency checks. We do
6497 : * it this way, rather than just making static arrays, for two reasons:
6498 : * (1) no need to waste the storage in most instantiations of the backend;
6499 : * (2) a static char array isn't guaranteed to have any particular
6500 : * alignment, whereas palloc() will provide MAXALIGN'd storage.
6501 : */
6502 1390 : replay_image_masked = (char *) palloc(BLCKSZ);
6503 1390 : master_image_masked = (char *) palloc(BLCKSZ);
6504 :
6505 1390 : if (read_backup_label(&checkPointLoc, &backupEndRequired,
6506 : &backupFromStandby))
6507 : {
6508 58 : List *tablespaces = NIL;
6509 :
6510 : /*
6511 : * Archive recovery was requested, and thanks to the backup label
6512 : * file, we know how far we need to replay to reach consistency. Enter
6513 : * archive recovery directly.
6514 : */
6515 58 : InArchiveRecovery = true;
6516 58 : if (StandbyModeRequested)
6517 56 : StandbyMode = true;
6518 :
6519 : /*
6520 : * When a backup_label file is present, we want to roll forward from
6521 : * the checkpoint it identifies, rather than using pg_control.
6522 : */
6523 58 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 0, true);
6524 58 : if (record != NULL)
6525 : {
6526 58 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
6527 58 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
6528 58 : ereport(DEBUG1,
6529 : (errmsg("checkpoint record is at %X/%X",
6530 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6531 58 : InRecovery = true; /* force recovery even if SHUTDOWNED */
6532 :
6533 : /*
6534 : * Make sure that REDO location exists. This may not be the case
6535 : * if there was a crash during an online backup, which left a
6536 : * backup_label around that references a WAL segment that's
6537 : * already been archived.
6538 : */
6539 58 : if (checkPoint.redo < checkPointLoc)
6540 : {
6541 58 : XLogBeginRead(xlogreader, checkPoint.redo);
6542 58 : if (!ReadRecord(xlogreader, LOG, false))
6543 0 : ereport(FATAL,
6544 : (errmsg("could not find redo location referenced by checkpoint record"),
6545 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" and add required recovery options.\n"
6546 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
6547 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
6548 : DataDir, DataDir, DataDir)));
6549 : }
6550 : }
6551 : else
6552 : {
6553 0 : ereport(FATAL,
6554 : (errmsg("could not locate required checkpoint record"),
6555 : errhint("If you are restoring from a backup, touch \"%s/recovery.signal\" and add required recovery options.\n"
6556 : "If you are not restoring from a backup, try removing the file \"%s/backup_label\".\n"
6557 : "Be careful: removing \"%s/backup_label\" will result in a corrupt cluster if restoring from a backup.",
6558 : DataDir, DataDir, DataDir)));
6559 : wasShutdown = false; /* keep compiler quiet */
6560 : }
6561 :
6562 : /* read the tablespace_map file if present and create symlinks. */
6563 58 : if (read_tablespace_map(&tablespaces))
6564 : {
6565 : ListCell *lc;
6566 :
6567 0 : foreach(lc, tablespaces)
6568 : {
6569 0 : tablespaceinfo *ti = lfirst(lc);
6570 : char *linkloc;
6571 :
6572 0 : linkloc = psprintf("pg_tblspc/%s", ti->oid);
6573 :
6574 : /*
6575 : * Remove the existing symlink if any and Create the symlink
6576 : * under PGDATA.
6577 : */
6578 0 : remove_tablespace_symlink(linkloc);
6579 :
6580 0 : if (symlink(ti->path, linkloc) < 0)
6581 0 : ereport(ERROR,
6582 : (errcode_for_file_access(),
6583 : errmsg("could not create symbolic link \"%s\": %m",
6584 : linkloc)));
6585 :
6586 0 : pfree(ti->oid);
6587 0 : pfree(ti->path);
6588 0 : pfree(ti);
6589 : }
6590 :
6591 : /* set flag to delete it later */
6592 0 : haveTblspcMap = true;
6593 : }
6594 :
6595 : /* set flag to delete it later */
6596 58 : haveBackupLabel = true;
6597 : }
6598 : else
6599 : {
6600 : /*
6601 : * If tablespace_map file is present without backup_label file, there
6602 : * is no use of such file. There is no harm in retaining it, but it
6603 : * is better to get rid of the map file so that we don't have any
6604 : * redundant file in data directory and it will avoid any sort of
6605 : * confusion. It seems prudent though to just rename the file out of
6606 : * the way rather than delete it completely, also we ignore any error
6607 : * that occurs in rename operation as even if map file is present
6608 : * without backup_label file, it is harmless.
6609 : */
6610 1332 : if (stat(TABLESPACE_MAP, &st) == 0)
6611 : {
6612 0 : unlink(TABLESPACE_MAP_OLD);
6613 0 : if (durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, DEBUG1) == 0)
6614 0 : ereport(LOG,
6615 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
6616 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
6617 : errdetail("File \"%s\" was renamed to \"%s\".",
6618 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
6619 : else
6620 0 : ereport(LOG,
6621 : (errmsg("ignoring file \"%s\" because no file \"%s\" exists",
6622 : TABLESPACE_MAP, BACKUP_LABEL_FILE),
6623 : errdetail("Could not rename file \"%s\" to \"%s\": %m.",
6624 : TABLESPACE_MAP, TABLESPACE_MAP_OLD)));
6625 : }
6626 :
6627 : /*
6628 : * It's possible that archive recovery was requested, but we don't
6629 : * know how far we need to replay the WAL before we reach consistency.
6630 : * This can happen for example if a base backup is taken from a
6631 : * running server using an atomic filesystem snapshot, without calling
6632 : * pg_start/stop_backup. Or if you just kill a running master server
6633 : * and put it into archive recovery by creating a recovery signal
6634 : * file.
6635 : *
6636 : * Our strategy in that case is to perform crash recovery first,
6637 : * replaying all the WAL present in pg_wal, and only enter archive
6638 : * recovery after that.
6639 : *
6640 : * But usually we already know how far we need to replay the WAL (up
6641 : * to minRecoveryPoint, up to backupEndPoint, or until we see an
6642 : * end-of-backup record), and we can enter archive recovery directly.
6643 : */
6644 1332 : if (ArchiveRecoveryRequested &&
6645 24 : (ControlFile->minRecoveryPoint != InvalidXLogRecPtr ||
6646 6 : ControlFile->backupEndRequired ||
6647 6 : ControlFile->backupEndPoint != InvalidXLogRecPtr ||
6648 6 : ControlFile->state == DB_SHUTDOWNED))
6649 : {
6650 22 : InArchiveRecovery = true;
6651 22 : if (StandbyModeRequested)
6652 22 : StandbyMode = true;
6653 : }
6654 :
6655 : /* Get the last valid checkpoint record. */
6656 1332 : checkPointLoc = ControlFile->checkPoint;
6657 1332 : RedoStartLSN = ControlFile->checkPointCopy.redo;
6658 1332 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 1, true);
6659 1332 : if (record != NULL)
6660 : {
6661 1332 : ereport(DEBUG1,
6662 : (errmsg("checkpoint record is at %X/%X",
6663 : (uint32) (checkPointLoc >> 32), (uint32) checkPointLoc)));
6664 : }
6665 : else
6666 : {
6667 : /*
6668 : * We used to attempt to go back to a secondary checkpoint record
6669 : * here, but only when not in standby mode. We now just fail if we
6670 : * can't read the last checkpoint because this allows us to
6671 : * simplify processing around checkpoints.
6672 : */
6673 0 : ereport(PANIC,
6674 : (errmsg("could not locate a valid checkpoint record")));
6675 : }
6676 1332 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
6677 1332 : wasShutdown = ((record->xl_info & ~XLR_INFO_MASK) == XLOG_CHECKPOINT_SHUTDOWN);
6678 : }
6679 :
6680 : /*
6681 : * Clear out any old relcache cache files. This is *necessary* if we do
6682 : * any WAL replay, since that would probably result in the cache files
6683 : * being out of sync with database reality. In theory we could leave them
6684 : * in place if the database had been cleanly shut down, but it seems
6685 : * safest to just remove them always and let them be rebuilt during the
6686 : * first backend startup. These files needs to be removed from all
6687 : * directories including pg_tblspc, however the symlinks are created only
6688 : * after reading tablespace_map file in case of archive recovery from
6689 : * backup, so needs to clear old relcache files here after creating
6690 : * symlinks.
6691 : */
6692 1390 : RelationCacheInitFileRemove();
6693 :
6694 : /*
6695 : * If the location of the checkpoint record is not on the expected
6696 : * timeline in the history of the requested timeline, we cannot proceed:
6697 : * the backup is not part of the history of the requested timeline.
6698 : */
6699 : Assert(expectedTLEs); /* was initialized by reading checkpoint
6700 : * record */
6701 1390 : if (tliOfPointInHistory(checkPointLoc, expectedTLEs) !=
6702 1390 : checkPoint.ThisTimeLineID)
6703 : {
6704 : XLogRecPtr switchpoint;
6705 :
6706 : /*
6707 : * tliSwitchPoint will throw an error if the checkpoint's timeline is
6708 : * not in expectedTLEs at all.
6709 : */
6710 0 : switchpoint = tliSwitchPoint(ControlFile->checkPointCopy.ThisTimeLineID, expectedTLEs, NULL);
6711 0 : ereport(FATAL,
6712 : (errmsg("requested timeline %u is not a child of this server's history",
6713 : recoveryTargetTLI),
6714 : 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.",
6715 : (uint32) (ControlFile->checkPoint >> 32),
6716 : (uint32) ControlFile->checkPoint,
6717 : ControlFile->checkPointCopy.ThisTimeLineID,
6718 : (uint32) (switchpoint >> 32),
6719 : (uint32) switchpoint)));
6720 : }
6721 :
6722 : /*
6723 : * The min recovery point should be part of the requested timeline's
6724 : * history, too.
6725 : */
6726 1390 : if (!XLogRecPtrIsInvalid(ControlFile->minRecoveryPoint) &&
6727 20 : tliOfPointInHistory(ControlFile->minRecoveryPoint - 1, expectedTLEs) !=
6728 20 : ControlFile->minRecoveryPointTLI)
6729 0 : ereport(FATAL,
6730 : (errmsg("requested timeline %u does not contain minimum recovery point %X/%X on timeline %u",
6731 : recoveryTargetTLI,
6732 : (uint32) (ControlFile->minRecoveryPoint >> 32),
6733 : (uint32) ControlFile->minRecoveryPoint,
6734 : ControlFile->minRecoveryPointTLI)));
6735 :
6736 1390 : LastRec = RecPtr = checkPointLoc;
6737 :
6738 1390 : ereport(DEBUG1,
6739 : (errmsg_internal("redo record is at %X/%X; shutdown %s",
6740 : (uint32) (checkPoint.redo >> 32), (uint32) checkPoint.redo,
6741 : wasShutdown ? "true" : "false")));
6742 1390 : ereport(DEBUG1,
6743 : (errmsg_internal("next transaction ID: " UINT64_FORMAT "; next OID: %u",
6744 : U64FromFullTransactionId(checkPoint.nextFullXid),
6745 : checkPoint.nextOid)));
6746 1390 : ereport(DEBUG1,
6747 : (errmsg_internal("next MultiXactId: %u; next MultiXactOffset: %u",
6748 : checkPoint.nextMulti, checkPoint.nextMultiOffset)));
6749 1390 : ereport(DEBUG1,
6750 : (errmsg_internal("oldest unfrozen transaction ID: %u, in database %u",
6751 : checkPoint.oldestXid, checkPoint.oldestXidDB)));
6752 1390 : ereport(DEBUG1,
6753 : (errmsg_internal("oldest MultiXactId: %u, in database %u",
6754 : checkPoint.oldestMulti, checkPoint.oldestMultiDB)));
6755 1390 : ereport(DEBUG1,
6756 : (errmsg_internal("commit timestamp Xid oldest/newest: %u/%u",
6757 : checkPoint.oldestCommitTsXid,
6758 : checkPoint.newestCommitTsXid)));
6759 1390 : if (!TransactionIdIsNormal(XidFromFullTransactionId(checkPoint.nextFullXid)))
6760 0 : ereport(PANIC,
6761 : (errmsg("invalid next transaction ID")));
6762 :
6763 : /* initialize shared memory variables from the checkpoint record */
6764 1390 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
6765 1390 : ShmemVariableCache->nextOid = checkPoint.nextOid;
6766 1390 : ShmemVariableCache->oidCount = 0;
6767 1390 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
6768 1390 : AdvanceOldestClogXid(checkPoint.oldestXid);
6769 1390 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
6770 1390 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
6771 1390 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
6772 : checkPoint.newestCommitTsXid);
6773 1390 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
6774 :
6775 : /*
6776 : * Initialize replication slots, before there's a chance to remove
6777 : * required resources.
6778 : */
6779 1390 : StartupReplicationSlots();
6780 :
6781 : /*
6782 : * Startup logical state, needs to be setup now so we have proper data
6783 : * during crash recovery.
6784 : */
6785 1390 : StartupReorderBuffer();
6786 :
6787 : /*
6788 : * Startup MultiXact. We need to do this early to be able to replay
6789 : * truncations.
6790 : */
6791 1390 : StartupMultiXact();
6792 :
6793 : /*
6794 : * Ditto for commit timestamps. Activate the facility if the setting is
6795 : * enabled in the control file, as there should be no tracking of commit
6796 : * timestamps done when the setting was disabled. This facility can be
6797 : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
6798 : */
6799 1390 : if (ControlFile->track_commit_timestamp)
6800 16 : StartupCommitTs();
6801 :
6802 : /*
6803 : * Recover knowledge about replay progress of known replication partners.
6804 : */
6805 1390 : StartupReplicationOrigin();
6806 :
6807 : /*
6808 : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
6809 : * control file. On recovery, all unlogged relations are blown away, so
6810 : * the unlogged LSN counter can be reset too.
6811 : */
6812 1390 : if (ControlFile->state == DB_SHUTDOWNED)
6813 1198 : XLogCtl->unloggedLSN = ControlFile->unloggedLSN;
6814 : else
6815 192 : XLogCtl->unloggedLSN = FirstNormalUnloggedLSN;
6816 :
6817 : /*
6818 : * We must replay WAL entries using the same TimeLineID they were created
6819 : * under, so temporarily adopt the TLI indicated by the checkpoint (see
6820 : * also xlog_redo()).
6821 : */
6822 1390 : ThisTimeLineID = checkPoint.ThisTimeLineID;
6823 :
6824 : /*
6825 : * Copy any missing timeline history files between 'now' and the recovery
6826 : * target timeline from archive to pg_wal. While we don't need those files
6827 : * ourselves - the history file of the recovery target timeline covers all
6828 : * the previous timelines in the history too - a cascading standby server
6829 : * might be interested in them. Or, if you archive the WAL from this
6830 : * server to a different archive than the master, it'd be good for all the
6831 : * history files to get archived there after failover, so that you can use
6832 : * one of the old timelines as a PITR target. Timeline history files are
6833 : * small, so it's better to copy them unnecessarily than not copy them and
6834 : * regret later.
6835 : */
6836 1390 : restoreTimeLineHistoryFiles(ThisTimeLineID, recoveryTargetTLI);
6837 :
6838 : /*
6839 : * Before running in recovery, scan pg_twophase and fill in its status to
6840 : * be able to work on entries generated by redo. Doing a scan before
6841 : * taking any recovery action has the merit to discard any 2PC files that
6842 : * are newer than the first record to replay, saving from any conflicts at
6843 : * replay. This avoids as well any subsequent scans when doing recovery
6844 : * of the on-disk two-phase data.
6845 : */
6846 1390 : restoreTwoPhaseData();
6847 :
6848 1390 : lastFullPageWrites = checkPoint.fullPageWrites;
6849 :
6850 1390 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
6851 1390 : doPageWrites = lastFullPageWrites;
6852 :
6853 1390 : if (RecPtr < checkPoint.redo)
6854 0 : ereport(PANIC,
6855 : (errmsg("invalid redo in checkpoint record")));
6856 :
6857 : /*
6858 : * Check whether we need to force recovery from WAL. If it appears to
6859 : * have been a clean shutdown and we did not have a recovery signal file,
6860 : * then assume no recovery needed.
6861 : */
6862 1390 : if (checkPoint.redo < RecPtr)
6863 : {
6864 100 : if (wasShutdown)
6865 0 : ereport(PANIC,
6866 : (errmsg("invalid redo record in shutdown checkpoint")));
6867 100 : InRecovery = true;
6868 : }
6869 1290 : else if (ControlFile->state != DB_SHUTDOWNED)
6870 92 : InRecovery = true;
6871 1198 : else if (ArchiveRecoveryRequested)
6872 : {
6873 : /* force recovery due to presence of recovery signal file */
6874 4 : InRecovery = true;
6875 : }
6876 :
6877 : /* REDO */
6878 1390 : if (InRecovery)
6879 : {
6880 : int rmid;
6881 :
6882 : /*
6883 : * Update pg_control to show that we are recovering and to show the
6884 : * selected checkpoint as the place we are starting from. We also mark
6885 : * pg_control with any minimum recovery stop point obtained from a
6886 : * backup history file.
6887 : */
6888 196 : dbstate_at_startup = ControlFile->state;
6889 196 : if (InArchiveRecovery)
6890 : {
6891 80 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6892 :
6893 80 : SpinLockAcquire(&XLogCtl->info_lck);
6894 80 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
6895 80 : SpinLockRelease(&XLogCtl->info_lck);
6896 : }
6897 : else
6898 : {
6899 116 : ereport(LOG,
6900 : (errmsg("database system was not properly shut down; "
6901 : "automatic recovery in progress")));
6902 116 : if (recoveryTargetTLI > ControlFile->checkPointCopy.ThisTimeLineID)
6903 2 : ereport(LOG,
6904 : (errmsg("crash recovery starts in timeline %u "
6905 : "and has target timeline %u",
6906 : ControlFile->checkPointCopy.ThisTimeLineID,
6907 : recoveryTargetTLI)));
6908 116 : ControlFile->state = DB_IN_CRASH_RECOVERY;
6909 :
6910 116 : SpinLockAcquire(&XLogCtl->info_lck);
6911 116 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
6912 116 : SpinLockRelease(&XLogCtl->info_lck);
6913 : }
6914 196 : ControlFile->checkPoint = checkPointLoc;
6915 196 : ControlFile->checkPointCopy = checkPoint;
6916 196 : if (InArchiveRecovery)
6917 : {
6918 : /* initialize minRecoveryPoint if not set yet */
6919 80 : if (ControlFile->minRecoveryPoint < checkPoint.redo)
6920 : {
6921 62 : ControlFile->minRecoveryPoint = checkPoint.redo;
6922 62 : ControlFile->minRecoveryPointTLI = checkPoint.ThisTimeLineID;
6923 : }
6924 : }
6925 :
6926 : /*
6927 : * Set backupStartPoint if we're starting recovery from a base backup.
6928 : *
6929 : * Also set backupEndPoint and use minRecoveryPoint as the backup end
6930 : * location if we're starting recovery from a base backup which was
6931 : * taken from a standby. In this case, the database system status in
6932 : * pg_control must indicate that the database was already in recovery.
6933 : * Usually that will be DB_IN_ARCHIVE_RECOVERY but also can be
6934 : * DB_SHUTDOWNED_IN_RECOVERY if recovery previously was interrupted
6935 : * before reaching this point; e.g. because restore_command or
6936 : * primary_conninfo were faulty.
6937 : *
6938 : * Any other state indicates that the backup somehow became corrupted
6939 : * and we can't sensibly continue with recovery.
6940 : */
6941 196 : if (haveBackupLabel)
6942 : {
6943 58 : ControlFile->backupStartPoint = checkPoint.redo;
6944 58 : ControlFile->backupEndRequired = backupEndRequired;
6945 :
6946 58 : if (backupFromStandby)
6947 : {
6948 0 : if (dbstate_at_startup != DB_IN_ARCHIVE_RECOVERY &&
6949 : dbstate_at_startup != DB_SHUTDOWNED_IN_RECOVERY)
6950 0 : ereport(FATAL,
6951 : (errmsg("backup_label contains data inconsistent with control file"),
6952 : errhint("This means that the backup is corrupted and you will "
6953 : "have to use another backup for recovery.")));
6954 0 : ControlFile->backupEndPoint = ControlFile->minRecoveryPoint;
6955 : }
6956 : }
6957 196 : ControlFile->time = (pg_time_t) time(NULL);
6958 : /* No need to hold ControlFileLock yet, we aren't up far enough */
6959 196 : UpdateControlFile();
6960 :
6961 : /*
6962 : * Initialize our local copy of minRecoveryPoint. When doing crash
6963 : * recovery we want to replay up to the end of WAL. Particularly, in
6964 : * the case of a promoted standby minRecoveryPoint value in the
6965 : * control file is only updated after the first checkpoint. However,
6966 : * if the instance crashes before the first post-recovery checkpoint
6967 : * is completed then recovery will use a stale location causing the
6968 : * startup process to think that there are still invalid page
6969 : * references when checking for data consistency.
6970 : */
6971 196 : if (InArchiveRecovery)
6972 : {
6973 80 : minRecoveryPoint = ControlFile->minRecoveryPoint;
6974 80 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6975 : }
6976 : else
6977 : {
6978 116 : minRecoveryPoint = InvalidXLogRecPtr;
6979 116 : minRecoveryPointTLI = 0;
6980 : }
6981 :
6982 : /*
6983 : * Reset pgstat data, because it may be invalid after recovery.
6984 : */
6985 196 : pgstat_reset_all();
6986 :
6987 : /*
6988 : * If there was a backup label file, it's done its job and the info
6989 : * has now been propagated into pg_control. We must get rid of the
6990 : * label file so that if we crash during recovery, we'll pick up at
6991 : * the latest recovery restartpoint instead of going all the way back
6992 : * to the backup start point. It seems prudent though to just rename
6993 : * the file out of the way rather than delete it completely.
6994 : */
6995 196 : if (haveBackupLabel)
6996 : {
6997 58 : unlink(BACKUP_LABEL_OLD);
6998 58 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
6999 : }
7000 :
7001 : /*
7002 : * If there was a tablespace_map file, it's done its job and the
7003 : * symlinks have been created. We must get rid of the map file so
7004 : * that if we crash during recovery, we don't create symlinks again.
7005 : * It seems prudent though to just rename the file out of the way
7006 : * rather than delete it completely.
7007 : */
7008 196 : if (haveTblspcMap)
7009 : {
7010 0 : unlink(TABLESPACE_MAP_OLD);
7011 0 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
7012 : }
7013 :
7014 : /* Check that the GUCs used to generate the WAL allow recovery */
7015 196 : CheckRequiredParameterValues();
7016 :
7017 : /*
7018 : * We're in recovery, so unlogged relations may be trashed and must be
7019 : * reset. This should be done BEFORE allowing Hot Standby
7020 : * connections, so that read-only backends don't try to read whatever
7021 : * garbage is left over from before.
7022 : */
7023 196 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
7024 :
7025 : /*
7026 : * Likewise, delete any saved transaction snapshot files that got left
7027 : * behind by crashed backends.
7028 : */
7029 196 : DeleteAllExportedSnapshotFiles();
7030 :
7031 : /*
7032 : * Initialize for Hot Standby, if enabled. We won't let backends in
7033 : * yet, not until we've reached the min recovery point specified in
7034 : * control file and we've established a recovery snapshot from a
7035 : * running-xacts WAL record.
7036 : */
7037 196 : if (ArchiveRecoveryRequested && EnableHotStandby)
7038 : {
7039 : TransactionId *xids;
7040 : int nxids;
7041 :
7042 82 : ereport(DEBUG1,
7043 : (errmsg("initializing for hot standby")));
7044 :
7045 82 : InitRecoveryTransactionEnvironment();
7046 :
7047 82 : if (wasShutdown)
7048 12 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
7049 : else
7050 70 : oldestActiveXID = checkPoint.oldestActiveXid;
7051 : Assert(TransactionIdIsValid(oldestActiveXID));
7052 :
7053 : /* Tell procarray about the range of xids it has to deal with */
7054 82 : ProcArrayInitRecovery(XidFromFullTransactionId(ShmemVariableCache->nextFullXid));
7055 :
7056 : /*
7057 : * Startup commit log and subtrans only. MultiXact and commit
7058 : * timestamp have already been started up and other SLRUs are not
7059 : * maintained during recovery and need not be started yet.
7060 : */
7061 82 : StartupCLOG();
7062 82 : StartupSUBTRANS(oldestActiveXID);
7063 :
7064 : /*
7065 : * If we're beginning at a shutdown checkpoint, we know that
7066 : * nothing was running on the master at this point. So fake-up an
7067 : * empty running-xacts record and use that here and now. Recover
7068 : * additional standby state for prepared transactions.
7069 : */
7070 82 : if (wasShutdown)
7071 : {
7072 : RunningTransactionsData running;
7073 : TransactionId latestCompletedXid;
7074 :
7075 : /*
7076 : * Construct a RunningTransactions snapshot representing a
7077 : * shut down server, with only prepared transactions still
7078 : * alive. We're never overflowed at this point because all
7079 : * subxids are listed with their parent prepared transactions.
7080 : */
7081 12 : running.xcnt = nxids;
7082 12 : running.subxcnt = 0;
7083 12 : running.subxid_overflow = false;
7084 12 : running.nextXid = XidFromFullTransactionId(checkPoint.nextFullXid);
7085 12 : running.oldestRunningXid = oldestActiveXID;
7086 12 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextFullXid);
7087 12 : TransactionIdRetreat(latestCompletedXid);
7088 : Assert(TransactionIdIsNormal(latestCompletedXid));
7089 12 : running.latestCompletedXid = latestCompletedXid;
7090 12 : running.xids = xids;
7091 :
7092 12 : ProcArrayApplyRecoveryInfo(&running);
7093 :
7094 12 : StandbyRecoverPreparedTransactions();
7095 : }
7096 : }
7097 :
7098 : /* Initialize resource managers */
7099 4508 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
7100 : {
7101 4312 : if (RmgrTable[rmid].rm_startup != NULL)
7102 784 : RmgrTable[rmid].rm_startup();
7103 : }
7104 :
7105 : /*
7106 : * Initialize shared variables for tracking progress of WAL replay, as
7107 : * if we had just replayed the record before the REDO location (or the
7108 : * checkpoint record itself, if it's a shutdown checkpoint).
7109 : */
7110 196 : SpinLockAcquire(&XLogCtl->info_lck);
7111 196 : if (checkPoint.redo < RecPtr)
7112 100 : XLogCtl->replayEndRecPtr = checkPoint.redo;
7113 : else
7114 96 : XLogCtl->replayEndRecPtr = EndRecPtr;
7115 196 : XLogCtl->replayEndTLI = ThisTimeLineID;
7116 196 : XLogCtl->lastReplayedEndRecPtr = XLogCtl->replayEndRecPtr;
7117 196 : XLogCtl->lastReplayedTLI = XLogCtl->replayEndTLI;
7118 196 : XLogCtl->recoveryLastXTime = 0;
7119 196 : XLogCtl->currentChunkStartTime = 0;
7120 196 : XLogCtl->recoveryPause = false;
7121 196 : SpinLockRelease(&XLogCtl->info_lck);
7122 :
7123 : /* Also ensure XLogReceiptTime has a sane value */
7124 196 : XLogReceiptTime = GetCurrentTimestamp();
7125 :
7126 : /*
7127 : * Let postmaster know we've started redo now, so that it can launch
7128 : * checkpointer to perform restartpoints. We don't bother during
7129 : * crash recovery as restartpoints can only be performed during
7130 : * archive recovery. And we'd like to keep crash recovery simple, to
7131 : * avoid introducing bugs that could affect you when recovering after
7132 : * crash.
7133 : *
7134 : * After this point, we can no longer assume that we're the only
7135 : * process in addition to postmaster! Also, fsync requests are
7136 : * subsequently to be handled by the checkpointer, not locally.
7137 : */
7138 196 : if (ArchiveRecoveryRequested && IsUnderPostmaster)
7139 : {
7140 82 : PublishStartupProcessInformation();
7141 82 : EnableSyncRequestForwarding();
7142 82 : SendPostmasterSignal(PMSIGNAL_RECOVERY_STARTED);
7143 82 : bgwriterLaunched = true;
7144 : }
7145 :
7146 : /*
7147 : * Allow read-only connections immediately if we're consistent
7148 : * already.
7149 : */
7150 196 : CheckRecoveryConsistency();
7151 :
7152 : /*
7153 : * Find the first record that logically follows the checkpoint --- it
7154 : * might physically precede it, though.
7155 : */
7156 196 : if (checkPoint.redo < RecPtr)
7157 : {
7158 : /* back up to find the record */
7159 100 : XLogBeginRead(xlogreader, checkPoint.redo);
7160 100 : record = ReadRecord(xlogreader, PANIC, false);
7161 : }
7162 : else
7163 : {
7164 : /* just have to read next record after CheckPoint */
7165 96 : record = ReadRecord(xlogreader, LOG, false);
7166 : }
7167 :
7168 196 : if (record != NULL)
7169 : {
7170 : ErrorContextCallback errcallback;
7171 : TimestampTz xtime;
7172 :
7173 188 : InRedo = true;
7174 :
7175 188 : ereport(LOG,
7176 : (errmsg("redo starts at %X/%X",
7177 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr)));
7178 :
7179 : /*
7180 : * main redo apply loop
7181 : */
7182 : do
7183 : {
7184 573454 : bool switchedTLI = false;
7185 :
7186 : #ifdef WAL_DEBUG
7187 : if (XLOG_DEBUG ||
7188 : (rmid == RM_XACT_ID && trace_recovery_messages <= DEBUG2) ||
7189 : (rmid != RM_XACT_ID && trace_recovery_messages <= DEBUG3))
7190 : {
7191 : StringInfoData buf;
7192 :
7193 : initStringInfo(&buf);
7194 : appendStringInfo(&buf, "REDO @ %X/%X; LSN %X/%X: ",
7195 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr,
7196 : (uint32) (EndRecPtr >> 32), (uint32) EndRecPtr);
7197 : xlog_outrec(&buf, xlogreader);
7198 : appendStringInfoString(&buf, " - ");
7199 : xlog_outdesc(&buf, xlogreader);
7200 : elog(LOG, "%s", buf.data);
7201 : pfree(buf.data);
7202 : }
7203 : #endif
7204 :
7205 : /* Handle interrupt signals of startup process */
7206 573454 : HandleStartupProcInterrupts();
7207 :
7208 : /*
7209 : * Pause WAL replay, if requested by a hot-standby session via
7210 : * SetRecoveryPause().
7211 : *
7212 : * Note that we intentionally don't take the info_lck spinlock
7213 : * here. We might therefore read a slightly stale value of
7214 : * the recoveryPause flag, but it can't be very stale (no
7215 : * worse than the last spinlock we did acquire). Since a
7216 : * pause request is a pretty asynchronous thing anyway,
7217 : * possibly responding to it one WAL record later than we
7218 : * otherwise would is a minor issue, so it doesn't seem worth
7219 : * adding another spinlock cycle to prevent that.
7220 : */
7221 573454 : if (((volatile XLogCtlData *) XLogCtl)->recoveryPause)
7222 0 : recoveryPausesHere(false);
7223 :
7224 : /*
7225 : * Have we reached our recovery target?
7226 : */
7227 573454 : if (recoveryStopsBefore(xlogreader))
7228 : {
7229 0 : reachedRecoveryTarget = true;
7230 0 : break;
7231 : }
7232 :
7233 : /*
7234 : * If we've been asked to lag the master, wait on latch until
7235 : * enough time has passed.
7236 : */
7237 573454 : if (recoveryApplyDelay(xlogreader))
7238 : {
7239 : /*
7240 : * We test for paused recovery again here. If user sets
7241 : * delayed apply, it may be because they expect to pause
7242 : * recovery in case of problems, so we must test again
7243 : * here otherwise pausing during the delay-wait wouldn't
7244 : * work.
7245 : */
7246 0 : if (((volatile XLogCtlData *) XLogCtl)->recoveryPause)
7247 0 : recoveryPausesHere(false);
7248 : }
7249 :
7250 : /* Setup error traceback support for ereport() */
7251 573454 : errcallback.callback = rm_redo_error_callback;
7252 573454 : errcallback.arg = (void *) xlogreader;
7253 573454 : errcallback.previous = error_context_stack;
7254 573454 : error_context_stack = &errcallback;
7255 :
7256 : /*
7257 : * ShmemVariableCache->nextFullXid must be beyond record's
7258 : * xid.
7259 : */
7260 573454 : AdvanceNextFullTransactionIdPastXid(record->xl_xid);
7261 :
7262 : /*
7263 : * Before replaying this record, check if this record causes
7264 : * the current timeline to change. The record is already
7265 : * considered to be part of the new timeline, so we update
7266 : * ThisTimeLineID before replaying it. That's important so
7267 : * that replayEndTLI, which is recorded as the minimum
7268 : * recovery point's TLI if recovery stops after this record,
7269 : * is set correctly.
7270 : */
7271 573454 : if (record->xl_rmid == RM_XLOG_ID)
7272 : {
7273 3040 : TimeLineID newTLI = ThisTimeLineID;
7274 3040 : TimeLineID prevTLI = ThisTimeLineID;
7275 3040 : uint8 info = record->xl_info & ~XLR_INFO_MASK;
7276 :
7277 3040 : if (info == XLOG_CHECKPOINT_SHUTDOWN)
7278 : {
7279 : CheckPoint checkPoint;
7280 :
7281 30 : memcpy(&checkPoint, XLogRecGetData(xlogreader), sizeof(CheckPoint));
7282 30 : newTLI = checkPoint.ThisTimeLineID;
7283 30 : prevTLI = checkPoint.PrevTimeLineID;
7284 : }
7285 3010 : else if (info == XLOG_END_OF_RECOVERY)
7286 : {
7287 : xl_end_of_recovery xlrec;
7288 :
7289 10 : memcpy(&xlrec, XLogRecGetData(xlogreader), sizeof(xl_end_of_recovery));
7290 10 : newTLI = xlrec.ThisTimeLineID;
7291 10 : prevTLI = xlrec.PrevTimeLineID;
7292 : }
7293 :
7294 3040 : if (newTLI != ThisTimeLineID)
7295 : {
7296 : /* Check that it's OK to switch to this TLI */
7297 10 : checkTimeLineSwitch(EndRecPtr, newTLI, prevTLI);
7298 :
7299 : /* Following WAL records should be run with new TLI */
7300 10 : ThisTimeLineID = newTLI;
7301 10 : switchedTLI = true;
7302 : }
7303 : }
7304 :
7305 : /*
7306 : * Update shared replayEndRecPtr before replaying this record,
7307 : * so that XLogFlush will update minRecoveryPoint correctly.
7308 : */
7309 573454 : SpinLockAcquire(&XLogCtl->info_lck);
7310 573454 : XLogCtl->replayEndRecPtr = EndRecPtr;
7311 573454 : XLogCtl->replayEndTLI = ThisTimeLineID;
7312 573454 : SpinLockRelease(&XLogCtl->info_lck);
7313 :
7314 : /*
7315 : * If we are attempting to enter Hot Standby mode, process
7316 : * XIDs we see
7317 : */
7318 573454 : if (standbyState >= STANDBY_INITIALIZED &&
7319 202592 : TransactionIdIsValid(record->xl_xid))
7320 201302 : RecordKnownAssignedTransactionIds(record->xl_xid);
7321 :
7322 : /* Now apply the WAL record itself */
7323 573454 : RmgrTable[record->xl_rmid].rm_redo(xlogreader);
7324 :
7325 : /*
7326 : * After redo, check whether the backup pages associated with
7327 : * the WAL record are consistent with the existing pages. This
7328 : * check is done only if consistency check is enabled for this
7329 : * record.
7330 : */
7331 573454 : if ((record->xl_info & XLR_CHECK_CONSISTENCY) != 0)
7332 0 : checkXLogConsistency(xlogreader);
7333 :
7334 : /* Pop the error context stack */
7335 573454 : error_context_stack = errcallback.previous;
7336 :
7337 : /*
7338 : * Update lastReplayedEndRecPtr after this record has been
7339 : * successfully replayed.
7340 : */
7341 573454 : SpinLockAcquire(&XLogCtl->info_lck);
7342 573454 : XLogCtl->lastReplayedEndRecPtr = EndRecPtr;
7343 573454 : XLogCtl->lastReplayedTLI = ThisTimeLineID;
7344 573454 : SpinLockRelease(&XLogCtl->info_lck);
7345 :
7346 : /*
7347 : * If rm_redo called XLogRequestWalReceiverReply, then we wake
7348 : * up the receiver so that it notices the updated
7349 : * lastReplayedEndRecPtr and sends a reply to the master.
7350 : */
7351 573454 : if (doRequestWalReceiverReply)
7352 : {
7353 0 : doRequestWalReceiverReply = false;
7354 0 : WalRcvForceReply();
7355 : }
7356 :
7357 : /* Remember this record as the last-applied one */
7358 573454 : LastRec = ReadRecPtr;
7359 :
7360 : /* Allow read-only connections if we're consistent now */
7361 573454 : CheckRecoveryConsistency();
7362 :
7363 : /* Is this a timeline switch? */
7364 573454 : if (switchedTLI)
7365 : {
7366 : /*
7367 : * Before we continue on the new timeline, clean up any
7368 : * (possibly bogus) future WAL segments on the old
7369 : * timeline.
7370 : */
7371 10 : RemoveNonParentXlogFiles(EndRecPtr, ThisTimeLineID);
7372 :
7373 : /*
7374 : * Wake up any walsenders to notice that we are on a new
7375 : * timeline.
7376 : */
7377 10 : if (switchedTLI && AllowCascadeReplication())
7378 10 : WalSndWakeup();
7379 : }
7380 :
7381 : /* Exit loop if we reached inclusive recovery target */
7382 573454 : if (recoveryStopsAfter(xlogreader))
7383 : {
7384 0 : reachedRecoveryTarget = true;
7385 0 : break;
7386 : }
7387 :
7388 : /* Else, try to fetch the next WAL record */
7389 573454 : record = ReadRecord(xlogreader, LOG, false);
7390 573422 : } while (record != NULL);
7391 :
7392 : /*
7393 : * end of main redo apply loop
7394 : */
7395 :
7396 156 : if (reachedRecoveryTarget)
7397 : {
7398 0 : if (!reachedConsistency)
7399 0 : ereport(FATAL,
7400 : (errmsg("requested recovery stop point is before consistent recovery point")));
7401 :
7402 : /*
7403 : * This is the last point where we can restart recovery with a
7404 : * new recovery target, if we shutdown and begin again. After
7405 : * this, Resource Managers may choose to do permanent
7406 : * corrective actions at end of recovery.
7407 : */
7408 0 : switch (recoveryTargetAction)
7409 : {
7410 0 : case RECOVERY_TARGET_ACTION_SHUTDOWN:
7411 :
7412 : /*
7413 : * exit with special return code to request shutdown
7414 : * of postmaster. Log messages issued from
7415 : * postmaster.
7416 : */
7417 0 : proc_exit(3);
7418 :
7419 0 : case RECOVERY_TARGET_ACTION_PAUSE:
7420 0 : SetRecoveryPause(true);
7421 0 : recoveryPausesHere(true);
7422 :
7423 : /* drop into promote */
7424 :
7425 0 : case RECOVERY_TARGET_ACTION_PROMOTE:
7426 0 : break;
7427 : }
7428 156 : }
7429 :
7430 : /* Allow resource managers to do any required cleanup. */
7431 3588 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
7432 : {
7433 3432 : if (RmgrTable[rmid].rm_cleanup != NULL)
7434 624 : RmgrTable[rmid].rm_cleanup();
7435 : }
7436 :
7437 156 : ereport(LOG,
7438 : (errmsg("redo done at %X/%X",
7439 : (uint32) (ReadRecPtr >> 32), (uint32) ReadRecPtr)));
7440 156 : xtime = GetLatestXTime();
7441 156 : if (xtime)
7442 38 : ereport(LOG,
7443 : (errmsg("last completed transaction was at log time %s",
7444 : timestamptz_to_str(xtime))));
7445 :
7446 156 : InRedo = false;
7447 : }
7448 : else
7449 : {
7450 : /* there are no WAL records following the checkpoint */
7451 8 : ereport(LOG,
7452 : (errmsg("redo is not required")));
7453 :
7454 : }
7455 :
7456 : /*
7457 : * This check is intentionally after the above log messages that
7458 : * indicate how far recovery went.
7459 : */
7460 164 : if (ArchiveRecoveryRequested &&
7461 50 : recoveryTarget != RECOVERY_TARGET_UNSET &&
7462 2 : !reachedRecoveryTarget)
7463 2 : ereport(FATAL,
7464 : (errmsg("recovery ended before configured recovery target was reached")));
7465 : }
7466 :
7467 : /*
7468 : * Kill WAL receiver, if it's still running, before we continue to write
7469 : * the startup checkpoint record. It will trump over the checkpoint and
7470 : * subsequent records if it's still alive when we start writing WAL.
7471 : */
7472 1356 : ShutdownWalRcv();
7473 :
7474 : /*
7475 : * Reset unlogged relations to the contents of their INIT fork. This is
7476 : * done AFTER recovery is complete so as to include any unlogged relations
7477 : * created during recovery, but BEFORE recovery is marked as having
7478 : * completed successfully. Otherwise we'd not retry if any of the post
7479 : * end-of-recovery steps fail.
7480 : */
7481 1356 : if (InRecovery)
7482 162 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
7483 :
7484 : /*
7485 : * We don't need the latch anymore. It's not strictly necessary to disown
7486 : * it, but let's do it for the sake of tidiness.
7487 : */
7488 1356 : if (ArchiveRecoveryRequested)
7489 48 : DisownLatch(&XLogCtl->recoveryWakeupLatch);
7490 :
7491 : /*
7492 : * We are now done reading the xlog from stream. Turn off streaming
7493 : * recovery to force fetching the files (which would be required at end of
7494 : * recovery, e.g., timeline history file) from archive or pg_wal.
7495 : *
7496 : * Note that standby mode must be turned off after killing WAL receiver,
7497 : * i.e., calling ShutdownWalRcv().
7498 : */
7499 : Assert(!WalRcvStreaming());
7500 1356 : StandbyMode = false;
7501 :
7502 : /*
7503 : * Re-fetch the last valid or last applied record, so we can identify the
7504 : * exact endpoint of what we consider the valid portion of WAL.
7505 : */
7506 1356 : XLogBeginRead(xlogreader, LastRec);
7507 1356 : record = ReadRecord(xlogreader, PANIC, false);
7508 1356 : EndOfLog = EndRecPtr;
7509 :
7510 : /*
7511 : * EndOfLogTLI is the TLI in the filename of the XLOG segment containing
7512 : * the end-of-log. It could be different from the timeline that EndOfLog
7513 : * nominally belongs to, if there was a timeline switch in that segment,
7514 : * and we were reading the old WAL from a segment belonging to a higher
7515 : * timeline.
7516 : */
7517 1356 : EndOfLogTLI = xlogreader->seg.ws_tli;
7518 :
7519 : /*
7520 : * Complain if we did not roll forward far enough to render the backup
7521 : * dump consistent. Note: it is indeed okay to look at the local variable
7522 : * minRecoveryPoint here, even though ControlFile->minRecoveryPoint might
7523 : * be further ahead --- ControlFile->minRecoveryPoint cannot have been
7524 : * advanced beyond the WAL we processed.
7525 : */
7526 1356 : if (InRecovery &&
7527 162 : (EndOfLog < minRecoveryPoint ||
7528 162 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint)))
7529 : {
7530 : /*
7531 : * Ran off end of WAL before reaching end-of-backup WAL record, or
7532 : * minRecoveryPoint. That's usually a bad sign, indicating that you
7533 : * tried to recover from an online backup but never called
7534 : * pg_stop_backup(), or you didn't archive all the WAL up to that
7535 : * point. However, this also happens in crash recovery, if the system
7536 : * crashes while an online backup is in progress. We must not treat
7537 : * that as an error, or the database will refuse to start up.
7538 : */
7539 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
7540 : {
7541 0 : if (ControlFile->backupEndRequired)
7542 0 : ereport(FATAL,
7543 : (errmsg("WAL ends before end of online backup"),
7544 : errhint("All WAL generated while online backup was taken must be available at recovery.")));
7545 0 : else if (!XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
7546 0 : ereport(FATAL,
7547 : (errmsg("WAL ends before end of online backup"),
7548 : 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.")));
7549 : else
7550 0 : ereport(FATAL,
7551 : (errmsg("WAL ends before consistent recovery point")));
7552 : }
7553 : }
7554 :
7555 : /*
7556 : * Pre-scan prepared transactions to find out the range of XIDs present.
7557 : * This information is not quite needed yet, but it is positioned here so
7558 : * as potential problems are detected before any on-disk change is done.
7559 : */
7560 1356 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
7561 :
7562 : /*
7563 : * Consider whether we need to assign a new timeline ID.
7564 : *
7565 : * If we are doing an archive recovery, we always assign a new ID. This
7566 : * handles a couple of issues. If we stopped short of the end of WAL
7567 : * during recovery, then we are clearly generating a new timeline and must
7568 : * assign it a unique new ID. Even if we ran to the end, modifying the
7569 : * current last segment is problematic because it may result in trying to
7570 : * overwrite an already-archived copy of that segment, and we encourage
7571 : * DBAs to make their archive_commands reject that. We can dodge the
7572 : * problem by making the new active segment have a new timeline ID.
7573 : *
7574 : * In a normal crash recovery, we can just extend the timeline we were in.
7575 : */
7576 1356 : PrevTimeLineID = ThisTimeLineID;
7577 1356 : if (ArchiveRecoveryRequested)
7578 : {
7579 : char reason[200];
7580 : char recoveryPath[MAXPGPATH];
7581 :
7582 : Assert(InArchiveRecovery);
7583 :
7584 48 : ThisTimeLineID = findNewestTimeLine(recoveryTargetTLI) + 1;
7585 48 : ereport(LOG,
7586 : (errmsg("selected new timeline ID: %u", ThisTimeLineID)));
7587 :
7588 : /*
7589 : * Create a comment for the history file to explain why and where
7590 : * timeline changed.
7591 : */
7592 48 : if (recoveryTarget == RECOVERY_TARGET_XID)
7593 0 : snprintf(reason, sizeof(reason),
7594 : "%s transaction %u",
7595 0 : recoveryStopAfter ? "after" : "before",
7596 : recoveryStopXid);
7597 48 : else if (recoveryTarget == RECOVERY_TARGET_TIME)
7598 0 : snprintf(reason, sizeof(reason),
7599 : "%s %s\n",
7600 0 : recoveryStopAfter ? "after" : "before",
7601 : timestamptz_to_str(recoveryStopTime));
7602 48 : else if (recoveryTarget == RECOVERY_TARGET_LSN)
7603 0 : snprintf(reason, sizeof(reason),
7604 : "%s LSN %X/%X\n",
7605 0 : recoveryStopAfter ? "after" : "before",
7606 0 : (uint32) (recoveryStopLSN >> 32),
7607 : (uint32) recoveryStopLSN);
7608 48 : else if (recoveryTarget == RECOVERY_TARGET_NAME)
7609 0 : snprintf(reason, sizeof(reason),
7610 : "at restore point \"%s\"",
7611 : recoveryStopName);
7612 48 : else if (recoveryTarget == RECOVERY_TARGET_IMMEDIATE)
7613 0 : snprintf(reason, sizeof(reason), "reached consistency");
7614 : else
7615 48 : snprintf(reason, sizeof(reason), "no recovery target specified");
7616 :
7617 : /*
7618 : * We are now done reading the old WAL. Turn off archive fetching if
7619 : * it was active, and make a writable copy of the last WAL segment.
7620 : * (Note that we also have a copy of the last block of the old WAL in
7621 : * readBuf; we will use that below.)
7622 : */
7623 48 : exitArchiveRecovery(EndOfLogTLI, EndOfLog);
7624 :
7625 : /*
7626 : * Write the timeline history file, and have it archived. After this
7627 : * point (or rather, as soon as the file is archived), the timeline
7628 : * will appear as "taken" in the WAL archive and to any standby
7629 : * servers. If we crash before actually switching to the new
7630 : * timeline, standby servers will nevertheless think that we switched
7631 : * to the new timeline, and will try to connect to the new timeline.
7632 : * To minimize the window for that, try to do as little as possible
7633 : * between here and writing the end-of-recovery record.
7634 : */
7635 48 : writeTimeLineHistory(ThisTimeLineID, recoveryTargetTLI,
7636 : EndRecPtr, reason);
7637 :
7638 : /*
7639 : * Since there might be a partial WAL segment named RECOVERYXLOG, get
7640 : * rid of it.
7641 : */
7642 48 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYXLOG");
7643 48 : unlink(recoveryPath); /* ignore any error */
7644 :
7645 : /* Get rid of any remaining recovered timeline-history file, too */
7646 48 : snprintf(recoveryPath, MAXPGPATH, XLOGDIR "/RECOVERYHISTORY");
7647 48 : unlink(recoveryPath); /* ignore any error */
7648 : }
7649 :
7650 : /* Save the selected TimeLineID in shared memory, too */
7651 1356 : XLogCtl->ThisTimeLineID = ThisTimeLineID;
7652 1356 : XLogCtl->PrevTimeLineID = PrevTimeLineID;
7653 :
7654 : /*
7655 : * Prepare to write WAL starting at EndOfLog location, and init xlog
7656 : * buffer cache using the block containing the last record from the
7657 : * previous incarnation.
7658 : */
7659 1356 : Insert = &XLogCtl->Insert;
7660 1356 : Insert->PrevBytePos = XLogRecPtrToBytePos(LastRec);
7661 1356 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
7662 :
7663 : /*
7664 : * Tricky point here: readBuf contains the *last* block that the LastRec
7665 : * record spans, not the one it starts in. The last block is indeed the
7666 : * one we want to use.
7667 : */
7668 1356 : if (EndOfLog % XLOG_BLCKSZ != 0)
7669 : {
7670 : char *page;
7671 : int len;
7672 : int firstIdx;
7673 : XLogRecPtr pageBeginPtr;
7674 :
7675 1338 : pageBeginPtr = EndOfLog - (EndOfLog % XLOG_BLCKSZ);
7676 : Assert(readOff == XLogSegmentOffset(pageBeginPtr, wal_segment_size));
7677 :
7678 1338 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
7679 :
7680 : /* Copy the valid part of the last block, and zero the rest */
7681 1338 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
7682 1338 : len = EndOfLog % XLOG_BLCKSZ;
7683 1338 : memcpy(page, xlogreader->readBuf, len);
7684 1338 : memset(page + len, 0, XLOG_BLCKSZ - len);
7685 :
7686 1338 : XLogCtl->xlblocks[firstIdx] = pageBeginPtr + XLOG_BLCKSZ;
7687 1338 : XLogCtl->InitializedUpTo = pageBeginPtr + XLOG_BLCKSZ;
7688 : }
7689 : else
7690 : {
7691 : /*
7692 : * There is no partial block to copy. Just set InitializedUpTo, and
7693 : * let the first attempt to insert a log record to initialize the next
7694 : * buffer.
7695 : */
7696 18 : XLogCtl->InitializedUpTo = EndOfLog;
7697 : }
7698 :
7699 1356 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
7700 :
7701 1356 : XLogCtl->LogwrtResult = LogwrtResult;
7702 :
7703 1356 : XLogCtl->LogwrtRqst.Write = EndOfLog;
7704 1356 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
7705 :
7706 : /*
7707 : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
7708 : * record before resource manager writes cleanup WAL records or checkpoint
7709 : * record is written.
7710 : */
7711 1356 : Insert->fullPageWrites = lastFullPageWrites;
7712 1356 : LocalSetXLogInsertAllowed();
7713 1356 : UpdateFullPageWrites();
7714 1356 : LocalXLogInsertAllowed = -1;
7715 :
7716 1356 : if (InRecovery)
7717 : {
7718 : /*
7719 : * Perform a checkpoint to update all our recovery activity to disk.
7720 : *
7721 : * Note that we write a shutdown checkpoint rather than an on-line
7722 : * one. This is not particularly critical, but since we may be
7723 : * assigning a new TLI, using a shutdown checkpoint allows us to have
7724 : * the rule that TLI only changes in shutdown checkpoints, which
7725 : * allows some extra error checking in xlog_redo.
7726 : *
7727 : * In fast promotion, only create a lightweight end-of-recovery record
7728 : * instead of a full checkpoint. A checkpoint is requested later,
7729 : * after we're fully out of recovery mode and already accepting
7730 : * queries.
7731 : */
7732 162 : if (bgwriterLaunched)
7733 : {
7734 48 : if (fast_promote)
7735 : {
7736 48 : checkPointLoc = ControlFile->checkPoint;
7737 :
7738 : /*
7739 : * Confirm the last checkpoint is available for us to recover
7740 : * from if we fail.
7741 : */
7742 48 : record = ReadCheckpointRecord(xlogreader, checkPointLoc, 1, false);
7743 48 : if (record != NULL)
7744 : {
7745 48 : fast_promoted = true;
7746 :
7747 : /*
7748 : * Insert a special WAL record to mark the end of
7749 : * recovery, since we aren't doing a checkpoint. That
7750 : * means that the checkpointer process may likely be in
7751 : * the middle of a time-smoothed restartpoint and could
7752 : * continue to be for minutes after this. That sounds
7753 : * strange, but the effect is roughly the same and it
7754 : * would be stranger to try to come out of the
7755 : * restartpoint and then checkpoint. We request a
7756 : * checkpoint later anyway, just for safety.
7757 : */
7758 48 : CreateEndOfRecoveryRecord();
7759 : }
7760 : }
7761 :
7762 48 : if (!fast_promoted)
7763 0 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
7764 : CHECKPOINT_IMMEDIATE |
7765 : CHECKPOINT_WAIT);
7766 : }
7767 : else
7768 114 : CreateCheckPoint(CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IMMEDIATE);
7769 : }
7770 :
7771 1356 : if (ArchiveRecoveryRequested)
7772 : {
7773 : /*
7774 : * And finally, execute the recovery_end_command, if any.
7775 : */
7776 48 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
7777 0 : ExecuteRecoveryCommand(recoveryEndCommand,
7778 : "recovery_end_command",
7779 : true);
7780 :
7781 : /*
7782 : * We switched to a new timeline. Clean up segments on the old
7783 : * timeline.
7784 : *
7785 : * If there are any higher-numbered segments on the old timeline,
7786 : * remove them. They might contain valid WAL, but they might also be
7787 : * pre-allocated files containing garbage. In any case, they are not
7788 : * part of the new timeline's history so we don't need them.
7789 : */
7790 48 : RemoveNonParentXlogFiles(EndOfLog, ThisTimeLineID);
7791 :
7792 : /*
7793 : * If the switch happened in the middle of a segment, what to do with
7794 : * the last, partial segment on the old timeline? If we don't archive
7795 : * it, and the server that created the WAL never archives it either
7796 : * (e.g. because it was hit by a meteor), it will never make it to the
7797 : * archive. That's OK from our point of view, because the new segment
7798 : * that we created with the new TLI contains all the WAL from the old
7799 : * timeline up to the switch point. But if you later try to do PITR to
7800 : * the "missing" WAL on the old timeline, recovery won't find it in
7801 : * the archive. It's physically present in the new file with new TLI,
7802 : * but recovery won't look there when it's recovering to the older
7803 : * timeline. On the other hand, if we archive the partial segment, and
7804 : * the original server on that timeline is still running and archives
7805 : * the completed version of the same segment later, it will fail. (We
7806 : * used to do that in 9.4 and below, and it caused such problems).
7807 : *
7808 : * As a compromise, we rename the last segment with the .partial
7809 : * suffix, and archive it. Archive recovery will never try to read
7810 : * .partial segments, so they will normally go unused. But in the odd
7811 : * PITR case, the administrator can copy them manually to the pg_wal
7812 : * directory (removing the suffix). They can be useful in debugging,
7813 : * too.
7814 : *
7815 : * If a .done or .ready file already exists for the old timeline,
7816 : * however, we had already determined that the segment is complete, so
7817 : * we can let it be archived normally. (In particular, if it was
7818 : * restored from the archive to begin with, it's expected to have a
7819 : * .done file).
7820 : */
7821 48 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
7822 : XLogArchivingActive())
7823 : {
7824 : char origfname[MAXFNAMELEN];
7825 : XLogSegNo endLogSegNo;
7826 :
7827 2 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
7828 2 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
7829 :
7830 2 : if (!XLogArchiveIsReadyOrDone(origfname))
7831 : {
7832 : char origpath[MAXPGPATH];
7833 : char partialfname[MAXFNAMELEN];
7834 : char partialpath[MAXPGPATH];
7835 :
7836 2 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
7837 2 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
7838 2 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
7839 :
7840 : /*
7841 : * Make sure there's no .done or .ready file for the .partial
7842 : * file.
7843 : */
7844 2 : XLogArchiveCleanup(partialfname);
7845 :
7846 2 : durable_rename(origpath, partialpath, ERROR);
7847 2 : XLogArchiveNotify(partialfname);
7848 : }
7849 : }
7850 : }
7851 :
7852 : /*
7853 : * Preallocate additional log files, if wanted.
7854 : */
7855 1356 : PreallocXlogFiles(EndOfLog);
7856 :
7857 : /*
7858 : * Okay, we're officially UP.
7859 : */
7860 1356 : InRecovery = false;
7861 :
7862 : /* start the archive_timeout timer and LSN running */
7863 1356 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
7864 1356 : XLogCtl->lastSegSwitchLSN = EndOfLog;
7865 :
7866 : /* also initialize latestCompletedXid, to nextXid - 1 */
7867 1356 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
7868 1356 : ShmemVariableCache->latestCompletedXid = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
7869 3504 : TransactionIdRetreat(ShmemVariableCache->latestCompletedXid);
7870 1356 : LWLockRelease(ProcArrayLock);
7871 :
7872 : /*
7873 : * Start up the commit log and subtrans, if not already done for hot
7874 : * standby. (commit timestamps are started below, if necessary.)
7875 : */
7876 1356 : if (standbyState == STANDBY_DISABLED)
7877 : {
7878 1308 : StartupCLOG();
7879 1308 : StartupSUBTRANS(oldestActiveXID);
7880 : }
7881 :
7882 : /*
7883 : * Perform end of recovery actions for any SLRUs that need it.
7884 : */
7885 1356 : TrimCLOG();
7886 1356 : TrimMultiXact();
7887 :
7888 : /* Reload shared-memory state for prepared transactions */
7889 1356 : RecoverPreparedTransactions();
7890 :
7891 : /*
7892 : * Shutdown the recovery environment. This must occur after
7893 : * RecoverPreparedTransactions(), see notes for lock_twophase_recover()
7894 : */
7895 1356 : if (standbyState != STANDBY_DISABLED)
7896 48 : ShutdownRecoveryTransactionEnvironment();
7897 :
7898 : /* Shut down xlogreader */
7899 1356 : if (readFile >= 0)
7900 : {
7901 1342 : close(readFile);
7902 1342 : readFile = -1;
7903 : }
7904 1356 : XLogReaderFree(xlogreader);
7905 :
7906 : /*
7907 : * If any of the critical GUCs have changed, log them before we allow
7908 : * backends to write WAL.
7909 : */
7910 1356 : LocalSetXLogInsertAllowed();
7911 1356 : XLogReportParameters();
7912 :
7913 : /*
7914 : * Local WAL inserts enabled, so it's time to finish initialization of
7915 : * commit timestamp.
7916 : */
7917 1356 : CompleteCommitTsInitialization();
7918 :
7919 : /*
7920 : * All done with end-of-recovery actions.
7921 : *
7922 : * Now allow backends to write WAL and update the control file status in
7923 : * consequence. The boolean flag allowing backends to write WAL is
7924 : * updated while holding ControlFileLock to prevent other backends to look
7925 : * at an inconsistent state of the control file in shared memory. There
7926 : * is still a small window during which backends can write WAL and the
7927 : * control file is still referring to a system not in DB_IN_PRODUCTION
7928 : * state while looking at the on-disk control file.
7929 : *
7930 : * Also, although the boolean flag to allow WAL is probably atomic in
7931 : * itself, we use the info_lck here to ensure that there are no race
7932 : * conditions concerning visibility of other recent updates to shared
7933 : * memory.
7934 : */
7935 1356 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7936 1356 : ControlFile->state = DB_IN_PRODUCTION;
7937 1356 : ControlFile->time = (pg_time_t) time(NULL);
7938 :
7939 1356 : SpinLockAcquire(&XLogCtl->info_lck);
7940 1356 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_DONE;
7941 1356 : SpinLockRelease(&XLogCtl->info_lck);
7942 :
7943 1356 : UpdateControlFile();
7944 1356 : LWLockRelease(ControlFileLock);
7945 :
7946 : /*
7947 : * If there were cascading standby servers connected to us, nudge any wal
7948 : * sender processes to notice that we've been promoted.
7949 : */
7950 1356 : WalSndWakeup();
7951 :
7952 : /*
7953 : * If this was a fast promotion, request an (online) checkpoint now. This
7954 : * isn't required for consistency, but the last restartpoint might be far
7955 : * back, and in case of a crash, recovering from it might take a longer
7956 : * than is appropriate now that we're not in standby mode anymore.
7957 : */
7958 1356 : if (fast_promoted)
7959 48 : RequestCheckpoint(CHECKPOINT_FORCE);
7960 1356 : }
7961 :
7962 : /*
7963 : * Checks if recovery has reached a consistent state. When consistency is
7964 : * reached and we have a valid starting standby snapshot, tell postmaster
7965 : * that it can start accepting read-only connections.
7966 : */
7967 : static void
7968 573652 : CheckRecoveryConsistency(void)
7969 : {
7970 : XLogRecPtr lastReplayedEndRecPtr;
7971 :
7972 : /*
7973 : * During crash recovery, we don't reach a consistent state until we've
7974 : * replayed all the WAL.
7975 : */
7976 573652 : if (XLogRecPtrIsInvalid(minRecoveryPoint))
7977 371008 : return;
7978 :
7979 : Assert(InArchiveRecovery);
7980 :
7981 : /*
7982 : * assume that we are called in the startup process, and hence don't need
7983 : * a lock to read lastReplayedEndRecPtr
7984 : */
7985 202644 : lastReplayedEndRecPtr = XLogCtl->lastReplayedEndRecPtr;
7986 :
7987 : /*
7988 : * Have we reached the point where our base backup was completed?
7989 : */
7990 202644 : if (!XLogRecPtrIsInvalid(ControlFile->backupEndPoint) &&
7991 0 : ControlFile->backupEndPoint <= lastReplayedEndRecPtr)
7992 : {
7993 : /*
7994 : * We have reached the end of base backup, as indicated by pg_control.
7995 : * The data on disk is now consistent. Reset backupStartPoint and
7996 : * backupEndPoint, and update minRecoveryPoint to make sure we don't
7997 : * allow starting up at an earlier point even if recovery is stopped
7998 : * and restarted soon after this.
7999 : */
8000 0 : elog(DEBUG1, "end of backup reached");
8001 :
8002 0 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8003 :
8004 0 : if (ControlFile->minRecoveryPoint < lastReplayedEndRecPtr)
8005 0 : ControlFile->minRecoveryPoint = lastReplayedEndRecPtr;
8006 :
8007 0 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
8008 0 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
8009 0 : ControlFile->backupEndRequired = false;
8010 0 : UpdateControlFile();
8011 :
8012 0 : LWLockRelease(ControlFileLock);
8013 : }
8014 :
8015 : /*
8016 : * Have we passed our safe starting point? Note that minRecoveryPoint is
8017 : * known to be incorrectly set if ControlFile->backupEndRequired, until
8018 : * the XLOG_BACKUP_END arrives to advise us of the correct
8019 : * minRecoveryPoint. All we know prior to that is that we're not
8020 : * consistent yet.
8021 : */
8022 202644 : if (!reachedConsistency && !ControlFile->backupEndRequired &&
8023 236 : minRecoveryPoint <= lastReplayedEndRecPtr &&
8024 88 : XLogRecPtrIsInvalid(ControlFile->backupStartPoint))
8025 : {
8026 : /*
8027 : * Check to see if the XLOG sequence contained any unresolved
8028 : * references to uninitialized pages.
8029 : */
8030 82 : XLogCheckInvalidPages();
8031 :
8032 82 : reachedConsistency = true;
8033 82 : ereport(LOG,
8034 : (errmsg("consistent recovery state reached at %X/%X",
8035 : (uint32) (lastReplayedEndRecPtr >> 32),
8036 : (uint32) lastReplayedEndRecPtr)));
8037 : }
8038 :
8039 : /*
8040 : * Have we got a valid starting snapshot that will allow queries to be
8041 : * run? If so, we can tell postmaster that the database is consistent now,
8042 : * enabling connections.
8043 : */
8044 202644 : if (standbyState == STANDBY_SNAPSHOT_READY &&
8045 202460 : !LocalHotStandbyActive &&
8046 82 : reachedConsistency &&
8047 : IsUnderPostmaster)
8048 : {
8049 82 : SpinLockAcquire(&XLogCtl->info_lck);
8050 82 : XLogCtl->SharedHotStandbyActive = true;
8051 82 : SpinLockRelease(&XLogCtl->info_lck);
8052 :
8053 82 : LocalHotStandbyActive = true;
8054 :
8055 82 : SendPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY);
8056 : }
8057 : }
8058 :
8059 : /*
8060 : * Is the system still in recovery?
8061 : *
8062 : * Unlike testing InRecovery, this works in any process that's connected to
8063 : * shared memory.
8064 : *
8065 : * As a side-effect, we initialize the local TimeLineID and RedoRecPtr
8066 : * variables the first time we see that recovery is finished.
8067 : */
8068 : bool
8069 73156092 : RecoveryInProgress(void)
8070 : {
8071 : /*
8072 : * We check shared state each time only until we leave recovery mode. We
8073 : * can't re-enter recovery, so there's no need to keep checking after the
8074 : * shared variable has once been seen false.
8075 : */
8076 73156092 : if (!LocalRecoveryInProgress)
8077 72675956 : return false;
8078 : else
8079 : {
8080 : /*
8081 : * use volatile pointer to make sure we make a fresh read of the
8082 : * shared variable.
8083 : */
8084 480136 : volatile XLogCtlData *xlogctl = XLogCtl;
8085 :
8086 480136 : LocalRecoveryInProgress = (xlogctl->SharedRecoveryState != RECOVERY_STATE_DONE);
8087 :
8088 : /*
8089 : * Initialize TimeLineID and RedoRecPtr when we discover that recovery
8090 : * is finished. InitPostgres() relies upon this behaviour to ensure
8091 : * that InitXLOGAccess() is called at backend startup. (If you change
8092 : * this, see also LocalSetXLogInsertAllowed.)
8093 : */
8094 480136 : if (!LocalRecoveryInProgress)
8095 : {
8096 : /*
8097 : * If we just exited recovery, make sure we read TimeLineID and
8098 : * RedoRecPtr after SharedRecoveryState (for machines with weak
8099 : * memory ordering).
8100 : */
8101 11796 : pg_memory_barrier();
8102 11796 : InitXLOGAccess();
8103 : }
8104 :
8105 : /*
8106 : * Note: We don't need a memory barrier when we're still in recovery.
8107 : * We might exit recovery immediately after return, so the caller
8108 : * can't rely on 'true' meaning that we're still in recovery anyway.
8109 : */
8110 :
8111 480136 : return LocalRecoveryInProgress;
8112 : }
8113 : }
8114 :
8115 : /*
8116 : * Returns current recovery state from shared memory.
8117 : *
8118 : * This returned state is kept consistent with the contents of the control
8119 : * file. See details about the possible values of RecoveryState in xlog.h.
8120 : */
8121 : RecoveryState
8122 8 : GetRecoveryState(void)
8123 : {
8124 : RecoveryState retval;
8125 :
8126 8 : SpinLockAcquire(&XLogCtl->info_lck);
8127 8 : retval = XLogCtl->SharedRecoveryState;
8128 8 : SpinLockRelease(&XLogCtl->info_lck);
8129 :
8130 8 : return retval;
8131 : }
8132 :
8133 : /*
8134 : * Is HotStandby active yet? This is only important in special backends
8135 : * since normal backends won't ever be able to connect until this returns
8136 : * true. Postmaster knows this by way of signal, not via shared memory.
8137 : *
8138 : * Unlike testing standbyState, this works in any process that's connected to
8139 : * shared memory. (And note that standbyState alone doesn't tell the truth
8140 : * anyway.)
8141 : */
8142 : bool
8143 142 : HotStandbyActive(void)
8144 : {
8145 : /*
8146 : * We check shared state each time only until Hot Standby is active. We
8147 : * can't de-activate Hot Standby, so there's no need to keep checking
8148 : * after the shared variable has once been seen true.
8149 : */
8150 142 : if (LocalHotStandbyActive)
8151 22 : return true;
8152 : else
8153 : {
8154 : /* spinlock is essential on machines with weak memory ordering! */
8155 120 : SpinLockAcquire(&XLogCtl->info_lck);
8156 120 : LocalHotStandbyActive = XLogCtl->SharedHotStandbyActive;
8157 120 : SpinLockRelease(&XLogCtl->info_lck);
8158 :
8159 120 : return LocalHotStandbyActive;
8160 : }
8161 : }
8162 :
8163 : /*
8164 : * Like HotStandbyActive(), but to be used only in WAL replay code,
8165 : * where we don't need to ask any other process what the state is.
8166 : */
8167 : bool
8168 0 : HotStandbyActiveInReplay(void)
8169 : {
8170 : Assert(AmStartupProcess() || !IsPostmasterEnvironment);
8171 0 : return LocalHotStandbyActive;
8172 : }
8173 :
8174 : /*
8175 : * Is this process allowed to insert new WAL records?
8176 : *
8177 : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
8178 : * But we also have provisions for forcing the result "true" or "false"
8179 : * within specific processes regardless of the global state.
8180 : */
8181 : bool
8182 63851768 : XLogInsertAllowed(void)
8183 : {
8184 : /*
8185 : * If value is "unconditionally true" or "unconditionally false", just
8186 : * return it. This provides the normal fast path once recovery is known
8187 : * done.
8188 : */
8189 63851768 : if (LocalXLogInsertAllowed >= 0)
8190 63841844 : return (bool) LocalXLogInsertAllowed;
8191 :
8192 : /*
8193 : * Else, must check to see if we're still in recovery.
8194 : */
8195 9924 : if (RecoveryInProgress())
8196 4094 : return false;
8197 :
8198 : /*
8199 : * On exit from recovery, reset to "unconditionally true", since there is
8200 : * no need to keep checking.
8201 : */
8202 5830 : LocalXLogInsertAllowed = 1;
8203 5830 : return true;
8204 : }
8205 :
8206 : /*
8207 : * Make XLogInsertAllowed() return true in the current process only.
8208 : *
8209 : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
8210 : * and even call LocalSetXLogInsertAllowed() again after that.
8211 : */
8212 : static void
8213 2874 : LocalSetXLogInsertAllowed(void)
8214 : {
8215 : Assert(LocalXLogInsertAllowed == -1);
8216 2874 : LocalXLogInsertAllowed = 1;
8217 :
8218 : /* Initialize as RecoveryInProgress() would do when switching state */
8219 2874 : InitXLOGAccess();
8220 2874 : }
8221 :
8222 : /*
8223 : * Subroutine to try to fetch and validate a prior checkpoint record.
8224 : *
8225 : * whichChkpt identifies the checkpoint (merely for reporting purposes).
8226 : * 1 for "primary", 0 for "other" (backup_label)
8227 : */
8228 : static XLogRecord *
8229 1438 : ReadCheckpointRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr,
8230 : int whichChkpt, bool report)
8231 : {
8232 : XLogRecord *record;
8233 : uint8 info;
8234 :
8235 1438 : if (!XRecOffIsValid(RecPtr))
8236 : {
8237 0 : if (!report)
8238 0 : return NULL;
8239 :
8240 0 : switch (whichChkpt)
8241 : {
8242 0 : case 1:
8243 0 : ereport(LOG,
8244 : (errmsg("invalid primary checkpoint link in control file")));
8245 0 : break;
8246 0 : default:
8247 0 : ereport(LOG,
8248 : (errmsg("invalid checkpoint link in backup_label file")));
8249 0 : break;
8250 : }
8251 0 : return NULL;
8252 : }
8253 :
8254 1438 : XLogBeginRead(xlogreader, RecPtr);
8255 1438 : record = ReadRecord(xlogreader, LOG, true);
8256 :
8257 1438 : if (record == NULL)
8258 : {
8259 0 : if (!report)
8260 0 : return NULL;
8261 :
8262 0 : switch (whichChkpt)
8263 : {
8264 0 : case 1:
8265 0 : ereport(LOG,
8266 : (errmsg("invalid primary checkpoint record")));
8267 0 : break;
8268 0 : default:
8269 0 : ereport(LOG,
8270 : (errmsg("invalid checkpoint record")));
8271 0 : break;
8272 : }
8273 0 : return NULL;
8274 : }
8275 1438 : if (record->xl_rmid != RM_XLOG_ID)
8276 : {
8277 0 : switch (whichChkpt)
8278 : {
8279 0 : case 1:
8280 0 : ereport(LOG,
8281 : (errmsg("invalid resource manager ID in primary checkpoint record")));
8282 0 : break;
8283 0 : default:
8284 0 : ereport(LOG,
8285 : (errmsg("invalid resource manager ID in checkpoint record")));
8286 0 : break;
8287 : }
8288 0 : return NULL;
8289 : }
8290 1438 : info = record->xl_info & ~XLR_INFO_MASK;
8291 1438 : if (info != XLOG_CHECKPOINT_SHUTDOWN &&
8292 : info != XLOG_CHECKPOINT_ONLINE)
8293 : {
8294 0 : switch (whichChkpt)
8295 : {
8296 0 : case 1:
8297 0 : ereport(LOG,
8298 : (errmsg("invalid xl_info in primary checkpoint record")));
8299 0 : break;
8300 0 : default:
8301 0 : ereport(LOG,
8302 : (errmsg("invalid xl_info in checkpoint record")));
8303 0 : break;
8304 : }
8305 0 : return NULL;
8306 : }
8307 1438 : if (record->xl_tot_len != SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(CheckPoint))
8308 : {
8309 0 : switch (whichChkpt)
8310 : {
8311 0 : case 1:
8312 0 : ereport(LOG,
8313 : (errmsg("invalid length of primary checkpoint record")));
8314 0 : break;
8315 0 : default:
8316 0 : ereport(LOG,
8317 : (errmsg("invalid length of checkpoint record")));
8318 0 : break;
8319 : }
8320 0 : return NULL;
8321 : }
8322 1438 : return record;
8323 : }
8324 :
8325 : /*
8326 : * This must be called in a backend process before creating WAL records
8327 : * (except in a standalone backend, which does StartupXLOG instead). We need
8328 : * to initialize the local copies of ThisTimeLineID and RedoRecPtr.
8329 : *
8330 : * Note: before Postgres 8.0, we went to some effort to keep the postmaster
8331 : * process's copies of ThisTimeLineID and RedoRecPtr valid too. This was
8332 : * unnecessary however, since the postmaster itself never touches XLOG anyway.
8333 : */
8334 : void
8335 15036 : InitXLOGAccess(void)
8336 : {
8337 15036 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8338 :
8339 : /* ThisTimeLineID doesn't change so we need no lock to copy it */
8340 15036 : ThisTimeLineID = XLogCtl->ThisTimeLineID;
8341 : Assert(ThisTimeLineID != 0 || IsBootstrapProcessingMode());
8342 :
8343 : /* set wal_segment_size */
8344 15036 : wal_segment_size = ControlFile->xlog_seg_size;
8345 :
8346 : /* Use GetRedoRecPtr to copy the RedoRecPtr safely */
8347 15036 : (void) GetRedoRecPtr();
8348 : /* Also update our copy of doPageWrites. */
8349 15036 : doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
8350 :
8351 : /* Also initialize the working areas for constructing WAL records */
8352 15036 : InitXLogInsert();
8353 15036 : }
8354 :
8355 : /*
8356 : * Return the current Redo pointer from shared memory.
8357 : *
8358 : * As a side-effect, the local RedoRecPtr copy is updated.
8359 : */
8360 : XLogRecPtr
8361 165034 : GetRedoRecPtr(void)
8362 : {
8363 : XLogRecPtr ptr;
8364 :
8365 : /*
8366 : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
8367 : * grabbed a WAL insertion lock to read the master copy, someone might
8368 : * update it just after we've released the lock.
8369 : */
8370 165034 : SpinLockAcquire(&XLogCtl->info_lck);
8371 165034 : ptr = XLogCtl->RedoRecPtr;
8372 165034 : SpinLockRelease(&XLogCtl->info_lck);
8373 :
8374 165034 : if (RedoRecPtr < ptr)
8375 11086 : RedoRecPtr = ptr;
8376 :
8377 165034 : return RedoRecPtr;
8378 : }
8379 :
8380 : /*
8381 : * Return information needed to decide whether a modified block needs a
8382 : * full-page image to be included in the WAL record.
8383 : *
8384 : * The returned values are cached copies from backend-private memory, and
8385 : * possibly out-of-date. XLogInsertRecord will re-check them against
8386 : * up-to-date values, while holding the WAL insert lock.
8387 : */
8388 : void
8389 29608684 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
8390 : {
8391 29608684 : *RedoRecPtr_p = RedoRecPtr;
8392 29608684 : *doPageWrites_p = doPageWrites;
8393 29608684 : }
8394 :
8395 : /*
8396 : * GetInsertRecPtr -- Returns the current insert position.
8397 : *
8398 : * NOTE: The value *actually* returned is the position of the last full
8399 : * xlog page. It lags behind the real insert position by at most 1 page.
8400 : * For that, we don't need to scan through WAL insertion locks, and an
8401 : * approximation is enough for the current usage of this function.
8402 : */
8403 : XLogRecPtr
8404 2102 : GetInsertRecPtr(void)
8405 : {
8406 : XLogRecPtr recptr;
8407 :
8408 2102 : SpinLockAcquire(&XLogCtl->info_lck);
8409 2102 : recptr = XLogCtl->LogwrtRqst.Write;
8410 2102 : SpinLockRelease(&XLogCtl->info_lck);
8411 :
8412 2102 : return recptr;
8413 : }
8414 :
8415 : /*
8416 : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
8417 : * position known to be fsync'd to disk.
8418 : */
8419 : XLogRecPtr
8420 159340 : GetFlushRecPtr(void)
8421 : {
8422 159340 : SpinLockAcquire(&XLogCtl->info_lck);
8423 159340 : LogwrtResult = XLogCtl->LogwrtResult;
8424 159340 : SpinLockRelease(&XLogCtl->info_lck);
8425 :
8426 159340 : return LogwrtResult.Flush;
8427 : }
8428 :
8429 : /*
8430 : * GetLastImportantRecPtr -- Returns the LSN of the last important record
8431 : * inserted. All records not explicitly marked as unimportant are considered
8432 : * important.
8433 : *
8434 : * The LSN is determined by computing the maximum of
8435 : * WALInsertLocks[i].lastImportantAt.
8436 : */
8437 : XLogRecPtr
8438 3204 : GetLastImportantRecPtr(void)
8439 : {
8440 3204 : XLogRecPtr res = InvalidXLogRecPtr;
8441 : int i;
8442 :
8443 28836 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
8444 : {
8445 : XLogRecPtr last_important;
8446 :
8447 : /*
8448 : * Need to take a lock to prevent torn reads of the LSN, which are
8449 : * possible on some of the supported platforms. WAL insert locks only
8450 : * support exclusive mode, so we have to use that.
8451 : */
8452 25632 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
8453 25632 : last_important = WALInsertLocks[i].l.lastImportantAt;
8454 25632 : LWLockRelease(&WALInsertLocks[i].l.lock);
8455 :
8456 25632 : if (res < last_important)
8457 3748 : res = last_important;
8458 : }
8459 :
8460 3204 : return res;
8461 : }
8462 :
8463 : /*
8464 : * Get the time and LSN of the last xlog segment switch
8465 : */
8466 : pg_time_t
8467 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
8468 : {
8469 : pg_time_t result;
8470 :
8471 : /* Need WALWriteLock, but shared lock is sufficient */
8472 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
8473 0 : result = XLogCtl->lastSegSwitchTime;
8474 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
8475 0 : LWLockRelease(WALWriteLock);
8476 :
8477 0 : return result;
8478 : }
8479 :
8480 : /*
8481 : * This must be called ONCE during postmaster or standalone-backend shutdown
8482 : */
8483 : void
8484 1126 : ShutdownXLOG(int code, Datum arg)
8485 : {
8486 : /*
8487 : * We should have an aux process resource owner to use, and we should not
8488 : * be in a transaction that's installed some other resowner.
8489 : */
8490 : Assert(AuxProcessResourceOwner != NULL);
8491 : Assert(CurrentResourceOwner == NULL ||
8492 : CurrentResourceOwner == AuxProcessResourceOwner);
8493 1126 : CurrentResourceOwner = AuxProcessResourceOwner;
8494 :
8495 : /* Don't be chatty in standalone mode */
8496 1126 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
8497 : (errmsg("shutting down")));
8498 :
8499 : /*
8500 : * Signal walsenders to move to stopping state.
8501 : */
8502 1126 : WalSndInitStopping();
8503 :
8504 : /*
8505 : * Wait for WAL senders to be in stopping state. This prevents commands
8506 : * from writing new WAL.
8507 : */
8508 1126 : WalSndWaitStopping();
8509 :
8510 1126 : if (RecoveryInProgress())
8511 32 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
8512 : else
8513 : {
8514 : /*
8515 : * If archiving is enabled, rotate the last XLOG file so that all the
8516 : * remaining records are archived (postmaster wakes up the archiver
8517 : * process one more time at the end of shutdown). The checkpoint
8518 : * record will go to the next XLOG file and won't be archived (yet).
8519 : */
8520 1094 : if (XLogArchivingActive() && XLogArchiveCommandSet())
8521 0 : RequestXLogSwitch(false);
8522 :
8523 1094 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE);
8524 : }
8525 1126 : ShutdownCLOG();
8526 1126 : ShutdownCommitTs();
8527 1126 : ShutdownSUBTRANS();
8528 1126 : ShutdownMultiXact();
8529 1126 : }
8530 :
8531 : /*
8532 : * Log start of a checkpoint.
8533 : */
8534 : static void
8535 480 : LogCheckpointStart(int flags, bool restartpoint)
8536 : {
8537 480 : elog(LOG, "%s starting:%s%s%s%s%s%s%s%s",
8538 : restartpoint ? "restartpoint" : "checkpoint",
8539 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
8540 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
8541 : (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
8542 : (flags & CHECKPOINT_FORCE) ? " force" : "",
8543 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
8544 : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
8545 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
8546 : (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "");
8547 480 : }
8548 :
8549 : /*
8550 : * Log end of a checkpoint.
8551 : */
8552 : static void
8553 3172 : LogCheckpointEnd(bool restartpoint)
8554 : {
8555 : long write_secs,
8556 : sync_secs,
8557 : total_secs,
8558 : longest_secs,
8559 : average_secs;
8560 : int write_usecs,
8561 : sync_usecs,
8562 : total_usecs,
8563 : longest_usecs,
8564 : average_usecs;
8565 : uint64 average_sync_time;
8566 :
8567 3172 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
8568 :
8569 3172 : TimestampDifference(CheckpointStats.ckpt_write_t,
8570 : CheckpointStats.ckpt_sync_t,
8571 : &write_secs, &write_usecs);
8572 :
8573 3172 : TimestampDifference(CheckpointStats.ckpt_sync_t,
8574 : CheckpointStats.ckpt_sync_end_t,
8575 : &sync_secs, &sync_usecs);
8576 :
8577 : /* Accumulate checkpoint timing summary data, in milliseconds. */
8578 6344 : BgWriterStats.m_checkpoint_write_time +=
8579 3172 : write_secs * 1000 + write_usecs / 1000;
8580 6344 : BgWriterStats.m_checkpoint_sync_time +=
8581 3172 : sync_secs * 1000 + sync_usecs / 1000;
8582 :
8583 : /*
8584 : * All of the published timing statistics are accounted for. Only
8585 : * continue if a log message is to be written.
8586 : */
8587 3172 : if (!log_checkpoints)
8588 2692 : return;
8589 :
8590 480 : TimestampDifference(CheckpointStats.ckpt_start_t,
8591 : CheckpointStats.ckpt_end_t,
8592 : &total_secs, &total_usecs);
8593 :
8594 : /*
8595 : * Timing values returned from CheckpointStats are in microseconds.
8596 : * Convert to the second plus microsecond form that TimestampDifference
8597 : * returns for homogeneous printing.
8598 : */
8599 480 : longest_secs = (long) (CheckpointStats.ckpt_longest_sync / 1000000);
8600 480 : longest_usecs = CheckpointStats.ckpt_longest_sync -
8601 : (uint64) longest_secs * 1000000;
8602 :
8603 480 : average_sync_time = 0;
8604 480 : if (CheckpointStats.ckpt_sync_rels > 0)
8605 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
8606 0 : CheckpointStats.ckpt_sync_rels;
8607 480 : average_secs = (long) (average_sync_time / 1000000);
8608 480 : average_usecs = average_sync_time - (uint64) average_secs * 1000000;
8609 :
8610 480 : elog(LOG, "%s complete: wrote %d buffers (%.1f%%); "
8611 : "%d WAL file(s) added, %d removed, %d recycled; "
8612 : "write=%ld.%03d s, sync=%ld.%03d s, total=%ld.%03d s; "
8613 : "sync files=%d, longest=%ld.%03d s, average=%ld.%03d s; "
8614 : "distance=%d kB, estimate=%d kB",
8615 : restartpoint ? "restartpoint" : "checkpoint",
8616 : CheckpointStats.ckpt_bufs_written,
8617 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
8618 : CheckpointStats.ckpt_segs_added,
8619 : CheckpointStats.ckpt_segs_removed,
8620 : CheckpointStats.ckpt_segs_recycled,
8621 : write_secs, write_usecs / 1000,
8622 : sync_secs, sync_usecs / 1000,
8623 : total_secs, total_usecs / 1000,
8624 : CheckpointStats.ckpt_sync_rels,
8625 : longest_secs, longest_usecs / 1000,
8626 : average_secs, average_usecs / 1000,
8627 : (int) (PrevCheckPointDistance / 1024.0),
8628 : (int) (CheckPointDistanceEstimate / 1024.0));
8629 : }
8630 :
8631 : /*
8632 : * Update the estimate of distance between checkpoints.
8633 : *
8634 : * The estimate is used to calculate the number of WAL segments to keep
8635 : * preallocated, see XLOGfileslop().
8636 : */
8637 : static void
8638 3172 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
8639 : {
8640 : /*
8641 : * To estimate the number of segments consumed between checkpoints, keep a
8642 : * moving average of the amount of WAL generated in previous checkpoint
8643 : * cycles. However, if the load is bursty, with quiet periods and busy
8644 : * periods, we want to cater for the peak load. So instead of a plain
8645 : * moving average, let the average decline slowly if the previous cycle
8646 : * used less WAL than estimated, but bump it up immediately if it used
8647 : * more.
8648 : *
8649 : * When checkpoints are triggered by max_wal_size, this should converge to
8650 : * CheckpointSegments * wal_segment_size,
8651 : *
8652 : * Note: This doesn't pay any attention to what caused the checkpoint.
8653 : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
8654 : * starting a base backup, are counted the same as those created
8655 : * automatically. The slow-decline will largely mask them out, if they are
8656 : * not frequent. If they are frequent, it seems reasonable to count them
8657 : * in as any others; if you issue a manual checkpoint every 5 minutes and
8658 : * never let a timed checkpoint happen, it makes sense to base the
8659 : * preallocation on that 5 minute interval rather than whatever
8660 : * checkpoint_timeout is set to.
8661 : */
8662 3172 : PrevCheckPointDistance = nbytes;
8663 3172 : if (CheckPointDistanceEstimate < nbytes)
8664 1408 : CheckPointDistanceEstimate = nbytes;
8665 : else
8666 1764 : CheckPointDistanceEstimate =
8667 1764 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
8668 3172 : }
8669 :
8670 : /*
8671 : * Perform a checkpoint --- either during shutdown, or on-the-fly
8672 : *
8673 : * flags is a bitwise OR of the following:
8674 : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
8675 : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
8676 : * CHECKPOINT_IMMEDIATE: finish the checkpoint ASAP,
8677 : * ignoring checkpoint_completion_target parameter.
8678 : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
8679 : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
8680 : * CHECKPOINT_END_OF_RECOVERY).
8681 : * CHECKPOINT_FLUSH_ALL: also flush buffers of unlogged tables.
8682 : *
8683 : * Note: flags contains other bits, of interest here only for logging purposes.
8684 : * In particular note that this routine is synchronous and does not pay
8685 : * attention to CHECKPOINT_WAIT.
8686 : *
8687 : * If !shutdown then we are writing an online checkpoint. This is a very special
8688 : * kind of operation and WAL record because the checkpoint action occurs over
8689 : * a period of time yet logically occurs at just a single LSN. The logical
8690 : * position of the WAL record (redo ptr) is the same or earlier than the
8691 : * physical position. When we replay WAL we locate the checkpoint via its
8692 : * physical position then read the redo ptr and actually start replay at the
8693 : * earlier logical position. Note that we don't write *anything* to WAL at
8694 : * the logical position, so that location could be any other kind of WAL record.
8695 : * All of this mechanism allows us to continue working while we checkpoint.
8696 : * As a result, timing of actions is critical here and be careful to note that
8697 : * this function will likely take minutes to execute on a busy system.
8698 : */
8699 : void
8700 3142 : CreateCheckPoint(int flags)
8701 : {
8702 : bool shutdown;
8703 : CheckPoint checkPoint;
8704 : XLogRecPtr recptr;
8705 : XLogSegNo _logSegNo;
8706 3142 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8707 : uint32 freespace;
8708 : XLogRecPtr PriorRedoPtr;
8709 : XLogRecPtr curInsert;
8710 : XLogRecPtr last_important_lsn;
8711 : VirtualTransactionId *vxids;
8712 : int nvxids;
8713 :
8714 : /*
8715 : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
8716 : * issued at a different time.
8717 : */
8718 3142 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
8719 1208 : shutdown = true;
8720 : else
8721 1934 : shutdown = false;
8722 :
8723 : /* sanity check */
8724 3142 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
8725 0 : elog(ERROR, "can't create a checkpoint during recovery");
8726 :
8727 : /*
8728 : * Initialize InitXLogInsert working areas before entering the critical
8729 : * section. Normally, this is done by the first call to
8730 : * RecoveryInProgress() or LocalSetXLogInsertAllowed(), but when creating
8731 : * an end-of-recovery checkpoint, the LocalSetXLogInsertAllowed call is
8732 : * done below in a critical section, and InitXLogInsert cannot be called
8733 : * in a critical section.
8734 : */
8735 3142 : InitXLogInsert();
8736 :
8737 : /*
8738 : * Acquire CheckpointLock to ensure only one checkpoint happens at a time.
8739 : * (This is just pro forma, since in the present system structure there is
8740 : * only one process that is allowed to issue checkpoints at any given
8741 : * time.)
8742 : */
8743 3142 : LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
8744 :
8745 : /*
8746 : * Prepare to accumulate statistics.
8747 : *
8748 : * Note: because it is possible for log_checkpoints to change while a
8749 : * checkpoint proceeds, we always accumulate stats, even if
8750 : * log_checkpoints is currently off.
8751 : */
8752 34562 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
8753 3142 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
8754 :
8755 : /*
8756 : * Use a critical section to force system panic if we have trouble.
8757 : */
8758 3142 : START_CRIT_SECTION();
8759 :
8760 3142 : if (shutdown)
8761 : {
8762 1208 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8763 1208 : ControlFile->state = DB_SHUTDOWNING;
8764 1208 : ControlFile->time = (pg_time_t) time(NULL);
8765 1208 : UpdateControlFile();
8766 1208 : LWLockRelease(ControlFileLock);
8767 : }
8768 :
8769 : /*
8770 : * Let smgr prepare for checkpoint; this has to happen before we determine
8771 : * the REDO pointer. Note that smgr must not do anything that'd have to
8772 : * be undone if we decide no checkpoint is needed.
8773 : */
8774 3142 : SyncPreCheckpoint();
8775 :
8776 : /* Begin filling in the checkpoint WAL record */
8777 37704 : MemSet(&checkPoint, 0, sizeof(checkPoint));
8778 3142 : checkPoint.time = (pg_time_t) time(NULL);
8779 :
8780 : /*
8781 : * For Hot Standby, derive the oldestActiveXid before we fix the redo
8782 : * pointer. This allows us to begin accumulating changes to assemble our
8783 : * starting snapshot of locks and transactions.
8784 : */
8785 3142 : if (!shutdown && XLogStandbyInfoActive())
8786 1896 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId();
8787 : else
8788 1246 : checkPoint.oldestActiveXid = InvalidTransactionId;
8789 :
8790 : /*
8791 : * Get location of last important record before acquiring insert locks (as
8792 : * GetLastImportantRecPtr() also locks WAL locks).
8793 : */
8794 3142 : last_important_lsn = GetLastImportantRecPtr();
8795 :
8796 : /*
8797 : * We must block concurrent insertions while examining insert state to
8798 : * determine the checkpoint REDO pointer.
8799 : */
8800 3142 : WALInsertLockAcquireExclusive();
8801 3142 : curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
8802 :
8803 : /*
8804 : * If this isn't a shutdown or forced checkpoint, and if there has been no
8805 : * WAL activity requiring a checkpoint, skip it. The idea here is to
8806 : * avoid inserting duplicate checkpoints when the system is idle.
8807 : */
8808 3142 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
8809 : CHECKPOINT_FORCE)) == 0)
8810 : {
8811 12 : if (last_important_lsn == ControlFile->checkPoint)
8812 : {
8813 0 : WALInsertLockRelease();
8814 0 : LWLockRelease(CheckpointLock);
8815 0 : END_CRIT_SECTION();
8816 0 : ereport(DEBUG1,
8817 : (errmsg("checkpoint skipped because system is idle")));
8818 0 : return;
8819 : }
8820 : }
8821 :
8822 : /*
8823 : * An end-of-recovery checkpoint is created before anyone is allowed to
8824 : * write WAL. To allow us to write the checkpoint record, temporarily
8825 : * enable XLogInsertAllowed. (This also ensures ThisTimeLineID is
8826 : * initialized, which we need here and in AdvanceXLInsertBuffer.)
8827 : */
8828 3142 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8829 114 : LocalSetXLogInsertAllowed();
8830 :
8831 3142 : checkPoint.ThisTimeLineID = ThisTimeLineID;
8832 3142 : if (flags & CHECKPOINT_END_OF_RECOVERY)
8833 114 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
8834 : else
8835 3028 : checkPoint.PrevTimeLineID = ThisTimeLineID;
8836 :
8837 3142 : checkPoint.fullPageWrites = Insert->fullPageWrites;
8838 :
8839 : /*
8840 : * Compute new REDO record ptr = location of next XLOG record.
8841 : *
8842 : * NB: this is NOT necessarily where the checkpoint record itself will be,
8843 : * since other backends may insert more XLOG records while we're off doing
8844 : * the buffer flush work. Those XLOG records are logically after the
8845 : * checkpoint, even though physically before it. Got that?
8846 : */
8847 3142 : freespace = INSERT_FREESPACE(curInsert);
8848 3142 : if (freespace == 0)
8849 : {
8850 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
8851 0 : curInsert += SizeOfXLogLongPHD;
8852 : else
8853 0 : curInsert += SizeOfXLogShortPHD;
8854 : }
8855 3142 : checkPoint.redo = curInsert;
8856 :
8857 : /*
8858 : * Here we update the shared RedoRecPtr for future XLogInsert calls; this
8859 : * must be done while holding all the insertion locks.
8860 : *
8861 : * Note: if we fail to complete the checkpoint, RedoRecPtr will be left
8862 : * pointing past where it really needs to point. This is okay; the only
8863 : * consequence is that XLogInsert might back up whole buffers that it
8864 : * didn't really need to. We can't postpone advancing RedoRecPtr because
8865 : * XLogInserts that happen while we are dumping buffers must assume that
8866 : * their buffer changes are not included in the checkpoint.
8867 : */
8868 3142 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
8869 :
8870 : /*
8871 : * Now we can release the WAL insertion locks, allowing other xacts to
8872 : * proceed while we are flushing disk buffers.
8873 : */
8874 3142 : WALInsertLockRelease();
8875 :
8876 : /* Update the info_lck-protected copy of RedoRecPtr as well */
8877 3142 : SpinLockAcquire(&XLogCtl->info_lck);
8878 3142 : XLogCtl->RedoRecPtr = checkPoint.redo;
8879 3142 : SpinLockRelease(&XLogCtl->info_lck);
8880 :
8881 : /*
8882 : * If enabled, log checkpoint start. We postpone this until now so as not
8883 : * to log anything if we decided to skip the checkpoint.
8884 : */
8885 3142 : if (log_checkpoints)
8886 462 : LogCheckpointStart(flags, false);
8887 :
8888 : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
8889 :
8890 : /*
8891 : * Get the other info we need for the checkpoint record.
8892 : *
8893 : * We don't need to save oldestClogXid in the checkpoint, it only matters
8894 : * for the short period in which clog is being truncated, and if we crash
8895 : * during that we'll redo the clog truncation and fix up oldestClogXid
8896 : * there.
8897 : */
8898 3142 : LWLockAcquire(XidGenLock, LW_SHARED);
8899 3142 : checkPoint.nextFullXid = ShmemVariableCache->nextFullXid;
8900 3142 : checkPoint.oldestXid = ShmemVariableCache->oldestXid;
8901 3142 : checkPoint.oldestXidDB = ShmemVariableCache->oldestXidDB;
8902 3142 : LWLockRelease(XidGenLock);
8903 :
8904 3142 : LWLockAcquire(CommitTsLock, LW_SHARED);
8905 3142 : checkPoint.oldestCommitTsXid = ShmemVariableCache->oldestCommitTsXid;
8906 3142 : checkPoint.newestCommitTsXid = ShmemVariableCache->newestCommitTsXid;
8907 3142 : LWLockRelease(CommitTsLock);
8908 :
8909 3142 : LWLockAcquire(OidGenLock, LW_SHARED);
8910 3142 : checkPoint.nextOid = ShmemVariableCache->nextOid;
8911 3142 : if (!shutdown)
8912 1934 : checkPoint.nextOid += ShmemVariableCache->oidCount;
8913 3142 : LWLockRelease(OidGenLock);
8914 :
8915 3142 : MultiXactGetCheckptMulti(shutdown,
8916 : &checkPoint.nextMulti,
8917 : &checkPoint.nextMultiOffset,
8918 : &checkPoint.oldestMulti,
8919 : &checkPoint.oldestMultiDB);
8920 :
8921 : /*
8922 : * Having constructed the checkpoint record, ensure all shmem disk buffers
8923 : * and commit-log buffers are flushed to disk.
8924 : *
8925 : * This I/O could fail for various reasons. If so, we will fail to
8926 : * complete the checkpoint, but there is no reason to force a system
8927 : * panic. Accordingly, exit critical section while doing it.
8928 : */
8929 3142 : END_CRIT_SECTION();
8930 :
8931 : /*
8932 : * In some cases there are groups of actions that must all occur on one
8933 : * side or the other of a checkpoint record. Before flushing the
8934 : * checkpoint record we must explicitly wait for any backend currently
8935 : * performing those groups of actions.
8936 : *
8937 : * One example is end of transaction, so we must wait for any transactions
8938 : * that are currently in commit critical sections. If an xact inserted
8939 : * its commit record into XLOG just before the REDO point, then a crash
8940 : * restart from the REDO point would not replay that record, which means
8941 : * that our flushing had better include the xact's update of pg_xact. So
8942 : * we wait till he's out of his commit critical section before proceeding.
8943 : * See notes in RecordTransactionCommit().
8944 : *
8945 : * Because we've already released the insertion locks, this test is a bit
8946 : * fuzzy: it is possible that we will wait for xacts we didn't really need
8947 : * to wait for. But the delay should be short and it seems better to make
8948 : * checkpoint take a bit longer than to hold off insertions longer than
8949 : * necessary. (In fact, the whole reason we have this issue is that xact.c
8950 : * does commit record XLOG insertion and clog update as two separate steps
8951 : * protected by different locks, but again that seems best on grounds of
8952 : * minimizing lock contention.)
8953 : *
8954 : * A transaction that has not yet set delayChkpt when we look cannot be at
8955 : * risk, since he's not inserted his commit record yet; and one that's
8956 : * already cleared it is not at risk either, since he's done fixing clog
8957 : * and we will correctly flush the update below. So we cannot miss any
8958 : * xacts we need to wait for.
8959 : */
8960 3142 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids);
8961 3142 : if (nvxids > 0)
8962 : {
8963 : do
8964 : {
8965 4 : pg_usleep(10000L); /* wait for 10 msec */
8966 4 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids));
8967 : }
8968 3142 : pfree(vxids);
8969 :
8970 3142 : CheckPointGuts(checkPoint.redo, flags);
8971 :
8972 : /*
8973 : * Take a snapshot of running transactions and write this to WAL. This
8974 : * allows us to reconstruct the state of running transactions during
8975 : * archive recovery, if required. Skip, if this info disabled.
8976 : *
8977 : * If we are shutting down, or Startup process is completing crash
8978 : * recovery we don't need to write running xact data.
8979 : */
8980 3142 : if (!shutdown && XLogStandbyInfoActive())
8981 1896 : LogStandbySnapshot();
8982 :
8983 3142 : START_CRIT_SECTION();
8984 :
8985 : /*
8986 : * Now insert the checkpoint record into XLOG.
8987 : */
8988 3142 : XLogBeginInsert();
8989 3142 : XLogRegisterData((char *) (&checkPoint), sizeof(checkPoint));
8990 3142 : recptr = XLogInsert(RM_XLOG_ID,
8991 : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
8992 : XLOG_CHECKPOINT_ONLINE);
8993 :
8994 3142 : XLogFlush(recptr);
8995 :
8996 : /*
8997 : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
8998 : * overwritten at next startup. No-one should even try, this just allows
8999 : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
9000 : * to just temporarily disable writing until the system has exited
9001 : * recovery.
9002 : */
9003 3142 : if (shutdown)
9004 : {
9005 1208 : if (flags & CHECKPOINT_END_OF_RECOVERY)
9006 114 : LocalXLogInsertAllowed = -1; /* return to "check" state */
9007 : else
9008 1094 : LocalXLogInsertAllowed = 0; /* never again write WAL */
9009 : }
9010 :
9011 : /*
9012 : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
9013 : * = end of actual checkpoint record.
9014 : */
9015 3142 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
9016 0 : ereport(PANIC,
9017 : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
9018 :
9019 : /*
9020 : * Remember the prior checkpoint's redo ptr for
9021 : * UpdateCheckPointDistanceEstimate()
9022 : */
9023 3142 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
9024 :
9025 : /*
9026 : * Update the control file.
9027 : */
9028 3142 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9029 3142 : if (shutdown)
9030 1208 : ControlFile->state = DB_SHUTDOWNED;
9031 3142 : ControlFile->checkPoint = ProcLastRecPtr;
9032 3142 : ControlFile->checkPointCopy = checkPoint;
9033 3142 : ControlFile->time = (pg_time_t) time(NULL);
9034 : /* crash recovery should always recover to the end of WAL */
9035 3142 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
9036 3142 : ControlFile->minRecoveryPointTLI = 0;
9037 :
9038 : /*
9039 : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
9040 : * unused on non-shutdown checkpoints, but seems useful to store it always
9041 : * for debugging purposes.
9042 : */
9043 3142 : SpinLockAcquire(&XLogCtl->ulsn_lck);
9044 3142 : ControlFile->unloggedLSN = XLogCtl->unloggedLSN;
9045 3142 : SpinLockRelease(&XLogCtl->ulsn_lck);
9046 :
9047 3142 : UpdateControlFile();
9048 3142 : LWLockRelease(ControlFileLock);
9049 :
9050 : /* Update shared-memory copy of checkpoint XID/epoch */
9051 3142 : SpinLockAcquire(&XLogCtl->info_lck);
9052 3142 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
9053 3142 : SpinLockRelease(&XLogCtl->info_lck);
9054 :
9055 : /*
9056 : * We are now done with critical updates; no need for system panic if we
9057 : * have trouble while fooling with old log segments.
9058 : */
9059 3142 : END_CRIT_SECTION();
9060 :
9061 : /*
9062 : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
9063 : */
9064 3142 : SyncPostCheckpoint();
9065 :
9066 : /*
9067 : * Update the average distance between checkpoints if the prior checkpoint
9068 : * exists.
9069 : */
9070 3142 : if (PriorRedoPtr != InvalidXLogRecPtr)
9071 3142 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
9072 :
9073 : /*
9074 : * Delete old log files, those no longer needed for last checkpoint to
9075 : * prevent the disk holding the xlog from growing full.
9076 : */
9077 3142 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
9078 3142 : KeepLogSeg(recptr, &_logSegNo);
9079 3142 : InvalidateObsoleteReplicationSlots(_logSegNo);
9080 3142 : _logSegNo--;
9081 3142 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr);
9082 :
9083 : /*
9084 : * Make more log segments if needed. (Do this after recycling old log
9085 : * segments, since that may supply some of the needed files.)
9086 : */
9087 3142 : if (!shutdown)
9088 1934 : PreallocXlogFiles(recptr);
9089 :
9090 : /*
9091 : * Truncate pg_subtrans if possible. We can throw away all data before
9092 : * the oldest XMIN of any running transaction. No future transaction will
9093 : * attempt to reference any pg_subtrans entry older than that (see Asserts
9094 : * in subtrans.c). During recovery, though, we mustn't do this because
9095 : * StartupSUBTRANS hasn't been called yet.
9096 : */
9097 3142 : if (!RecoveryInProgress())
9098 3028 : TruncateSUBTRANS(GetOldestXmin(NULL, PROCARRAY_FLAGS_DEFAULT));
9099 :
9100 : /* Real work is done, but log and update stats before releasing lock. */
9101 3142 : LogCheckpointEnd(false);
9102 :
9103 : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
9104 : NBuffers,
9105 : CheckpointStats.ckpt_segs_added,
9106 : CheckpointStats.ckpt_segs_removed,
9107 : CheckpointStats.ckpt_segs_recycled);
9108 :
9109 3142 : LWLockRelease(CheckpointLock);
9110 : }
9111 :
9112 : /*
9113 : * Mark the end of recovery in WAL though without running a full checkpoint.
9114 : * We can expect that a restartpoint is likely to be in progress as we
9115 : * do this, though we are unwilling to wait for it to complete. So be
9116 : * careful to avoid taking the CheckpointLock anywhere here.
9117 : *
9118 : * CreateRestartPoint() allows for the case where recovery may end before
9119 : * the restartpoint completes so there is no concern of concurrent behaviour.
9120 : */
9121 : static void
9122 48 : CreateEndOfRecoveryRecord(void)
9123 : {
9124 : xl_end_of_recovery xlrec;
9125 : XLogRecPtr recptr;
9126 :
9127 : /* sanity check */
9128 48 : if (!RecoveryInProgress())
9129 0 : elog(ERROR, "can only be used to end recovery");
9130 :
9131 48 : xlrec.end_time = GetCurrentTimestamp();
9132 :
9133 48 : WALInsertLockAcquireExclusive();
9134 48 : xlrec.ThisTimeLineID = ThisTimeLineID;
9135 48 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
9136 48 : WALInsertLockRelease();
9137 :
9138 48 : LocalSetXLogInsertAllowed();
9139 :
9140 48 : START_CRIT_SECTION();
9141 :
9142 48 : XLogBeginInsert();
9143 48 : XLogRegisterData((char *) &xlrec, sizeof(xl_end_of_recovery));
9144 48 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
9145 :
9146 48 : XLogFlush(recptr);
9147 :
9148 : /*
9149 : * Update the control file so that crash recovery can follow the timeline
9150 : * changes to this point.
9151 : */
9152 48 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9153 48 : ControlFile->time = (pg_time_t) time(NULL);
9154 48 : ControlFile->minRecoveryPoint = recptr;
9155 48 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
9156 48 : UpdateControlFile();
9157 48 : LWLockRelease(ControlFileLock);
9158 :
9159 48 : END_CRIT_SECTION();
9160 :
9161 48 : LocalXLogInsertAllowed = -1; /* return to "check" state */
9162 48 : }
9163 :
9164 : /*
9165 : * Flush all data in shared memory to disk, and fsync
9166 : *
9167 : * This is the common code shared between regular checkpoints and
9168 : * recovery restartpoints.
9169 : */
9170 : static void
9171 3172 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
9172 : {
9173 3172 : CheckPointCLOG();
9174 3172 : CheckPointCommitTs();
9175 3172 : CheckPointSUBTRANS();
9176 3172 : CheckPointMultiXact();
9177 3172 : CheckPointPredicate();
9178 3172 : CheckPointRelationMap();
9179 3172 : CheckPointReplicationSlots();
9180 3172 : CheckPointSnapBuild();
9181 3172 : CheckPointLogicalRewriteHeap();
9182 3172 : CheckPointBuffers(flags); /* performs all required fsyncs */
9183 3172 : CheckPointReplicationOrigin();
9184 : /* We deliberately delay 2PC checkpointing as long as possible */
9185 3172 : CheckPointTwoPhase(checkPointRedo);
9186 3172 : }
9187 :
9188 : /*
9189 : * Save a checkpoint for recovery restart if appropriate
9190 : *
9191 : * This function is called each time a checkpoint record is read from XLOG.
9192 : * It must determine whether the checkpoint represents a safe restartpoint or
9193 : * not. If so, the checkpoint record is stashed in shared memory so that
9194 : * CreateRestartPoint can consult it. (Note that the latter function is
9195 : * executed by the checkpointer, while this one will be executed by the
9196 : * startup process.)
9197 : */
9198 : static void
9199 202 : RecoveryRestartPoint(const CheckPoint *checkPoint)
9200 : {
9201 : /*
9202 : * Also refrain from creating a restartpoint if we have seen any
9203 : * references to non-existent pages. Restarting recovery from the
9204 : * restartpoint would not see the references, so we would lose the
9205 : * cross-check that the pages belonged to a relation that was dropped
9206 : * later.
9207 : */
9208 202 : if (XLogHaveInvalidPages())
9209 : {
9210 0 : elog(trace_recovery(DEBUG2),
9211 : "could not record restart point at %X/%X because there "
9212 : "are unresolved references to invalid pages",
9213 : (uint32) (checkPoint->redo >> 32),
9214 : (uint32) checkPoint->redo);
9215 0 : return;
9216 : }
9217 :
9218 : /*
9219 : * Copy the checkpoint record to shared memory, so that checkpointer can
9220 : * work out the next time it wants to perform a restartpoint.
9221 : */
9222 202 : SpinLockAcquire(&XLogCtl->info_lck);
9223 202 : XLogCtl->lastCheckPointRecPtr = ReadRecPtr;
9224 202 : XLogCtl->lastCheckPointEndPtr = EndRecPtr;
9225 202 : XLogCtl->lastCheckPoint = *checkPoint;
9226 202 : SpinLockRelease(&XLogCtl->info_lck);
9227 : }
9228 :
9229 : /*
9230 : * Establish a restartpoint if possible.
9231 : *
9232 : * This is similar to CreateCheckPoint, but is used during WAL recovery
9233 : * to establish a point from which recovery can roll forward without
9234 : * replaying the entire recovery log.
9235 : *
9236 : * Returns true if a new restartpoint was established. We can only establish
9237 : * a restartpoint if we have replayed a safe checkpoint record since last
9238 : * restartpoint.
9239 : */
9240 : bool
9241 76 : CreateRestartPoint(int flags)
9242 : {
9243 : XLogRecPtr lastCheckPointRecPtr;
9244 : XLogRecPtr lastCheckPointEndPtr;
9245 : CheckPoint lastCheckPoint;
9246 : XLogRecPtr PriorRedoPtr;
9247 : XLogRecPtr receivePtr;
9248 : XLogRecPtr replayPtr;
9249 : TimeLineID replayTLI;
9250 : XLogRecPtr endptr;
9251 : XLogSegNo _logSegNo;
9252 : TimestampTz xtime;
9253 :
9254 : /*
9255 : * Acquire CheckpointLock to ensure only one restartpoint or checkpoint
9256 : * happens at a time.
9257 : */
9258 76 : LWLockAcquire(CheckpointLock, LW_EXCLUSIVE);
9259 :
9260 : /* Get a local copy of the last safe checkpoint record. */
9261 76 : SpinLockAcquire(&XLogCtl->info_lck);
9262 76 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
9263 76 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
9264 76 : lastCheckPoint = XLogCtl->lastCheckPoint;
9265 76 : SpinLockRelease(&XLogCtl->info_lck);
9266 :
9267 : /*
9268 : * Check that we're still in recovery mode. It's ok if we exit recovery
9269 : * mode after this check, the restart point is valid anyway.
9270 : */
9271 76 : if (!RecoveryInProgress())
9272 : {
9273 0 : ereport(DEBUG2,
9274 : (errmsg("skipping restartpoint, recovery has already ended")));
9275 0 : LWLockRelease(CheckpointLock);
9276 0 : return false;
9277 : }
9278 :
9279 : /*
9280 : * If the last checkpoint record we've replayed is already our last
9281 : * restartpoint, we can't perform a new restart point. We still update
9282 : * minRecoveryPoint in that case, so that if this is a shutdown restart
9283 : * point, we won't start up earlier than before. That's not strictly
9284 : * necessary, but when hot standby is enabled, it would be rather weird if
9285 : * the database opened up for read-only connections at a point-in-time
9286 : * before the last shutdown. Such time travel is still possible in case of
9287 : * immediate shutdown, though.
9288 : *
9289 : * We don't explicitly advance minRecoveryPoint when we do create a
9290 : * restartpoint. It's assumed that flushing the buffers will do that as a
9291 : * side-effect.
9292 : */
9293 76 : if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
9294 74 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
9295 : {
9296 46 : ereport(DEBUG2,
9297 : (errmsg("skipping restartpoint, already performed at %X/%X",
9298 : (uint32) (lastCheckPoint.redo >> 32),
9299 : (uint32) lastCheckPoint.redo)));
9300 :
9301 46 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
9302 46 : if (flags & CHECKPOINT_IS_SHUTDOWN)
9303 : {
9304 24 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9305 24 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
9306 24 : ControlFile->time = (pg_time_t) time(NULL);
9307 24 : UpdateControlFile();
9308 24 : LWLockRelease(ControlFileLock);
9309 : }
9310 46 : LWLockRelease(CheckpointLock);
9311 46 : return false;
9312 : }
9313 :
9314 : /*
9315 : * Update the shared RedoRecPtr so that the startup process can calculate
9316 : * the number of segments replayed since last restartpoint, and request a
9317 : * restartpoint if it exceeds CheckPointSegments.
9318 : *
9319 : * Like in CreateCheckPoint(), hold off insertions to update it, although
9320 : * during recovery this is just pro forma, because no WAL insertions are
9321 : * happening.
9322 : */
9323 30 : WALInsertLockAcquireExclusive();
9324 30 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
9325 30 : WALInsertLockRelease();
9326 :
9327 : /* Also update the info_lck-protected copy */
9328 30 : SpinLockAcquire(&XLogCtl->info_lck);
9329 30 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
9330 30 : SpinLockRelease(&XLogCtl->info_lck);
9331 :
9332 : /*
9333 : * Prepare to accumulate statistics.
9334 : *
9335 : * Note: because it is possible for log_checkpoints to change while a
9336 : * checkpoint proceeds, we always accumulate stats, even if
9337 : * log_checkpoints is currently off.
9338 : */
9339 330 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
9340 30 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
9341 :
9342 30 : if (log_checkpoints)
9343 18 : LogCheckpointStart(flags, true);
9344 :
9345 30 : CheckPointGuts(lastCheckPoint.redo, flags);
9346 :
9347 : /*
9348 : * Remember the prior checkpoint's redo ptr for
9349 : * UpdateCheckPointDistanceEstimate()
9350 : */
9351 30 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
9352 :
9353 : /*
9354 : * Update pg_control, using current time. Check that it still shows
9355 : * DB_IN_ARCHIVE_RECOVERY state and an older checkpoint, else do nothing;
9356 : * this is a quick hack to make sure nothing really bad happens if somehow
9357 : * we get here after the end-of-recovery checkpoint.
9358 : */
9359 30 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9360 30 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY &&
9361 30 : ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
9362 : {
9363 30 : ControlFile->checkPoint = lastCheckPointRecPtr;
9364 30 : ControlFile->checkPointCopy = lastCheckPoint;
9365 30 : ControlFile->time = (pg_time_t) time(NULL);
9366 :
9367 : /*
9368 : * Ensure minRecoveryPoint is past the checkpoint record. Normally,
9369 : * this will have happened already while writing out dirty buffers,
9370 : * but not necessarily - e.g. because no buffers were dirtied. We do
9371 : * this because a non-exclusive base backup uses minRecoveryPoint to
9372 : * determine which WAL files must be included in the backup, and the
9373 : * file (or files) containing the checkpoint record must be included,
9374 : * at a minimum. Note that for an ordinary restart of recovery there's
9375 : * no value in having the minimum recovery point any earlier than this
9376 : * anyway, because redo will begin just after the checkpoint record.
9377 : */
9378 30 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
9379 : {
9380 8 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
9381 8 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
9382 :
9383 : /* update local copy */
9384 8 : minRecoveryPoint = ControlFile->minRecoveryPoint;
9385 8 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
9386 : }
9387 30 : if (flags & CHECKPOINT_IS_SHUTDOWN)
9388 8 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
9389 30 : UpdateControlFile();
9390 : }
9391 30 : LWLockRelease(ControlFileLock);
9392 :
9393 : /*
9394 : * Update the average distance between checkpoints/restartpoints if the
9395 : * prior checkpoint exists.
9396 : */
9397 30 : if (PriorRedoPtr != InvalidXLogRecPtr)
9398 30 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
9399 :
9400 : /*
9401 : * Delete old log files, those no longer needed for last restartpoint to
9402 : * prevent the disk holding the xlog from growing full.
9403 : */
9404 30 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
9405 :
9406 : /*
9407 : * Retreat _logSegNo using the current end of xlog replayed or received,
9408 : * whichever is later.
9409 : */
9410 30 : receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
9411 30 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
9412 30 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
9413 30 : KeepLogSeg(endptr, &_logSegNo);
9414 30 : InvalidateObsoleteReplicationSlots(_logSegNo);
9415 30 : _logSegNo--;
9416 :
9417 : /*
9418 : * Try to recycle segments on a useful timeline. If we've been promoted
9419 : * since the beginning of this restartpoint, use the new timeline chosen
9420 : * at end of recovery (RecoveryInProgress() sets ThisTimeLineID in that
9421 : * case). If we're still in recovery, use the timeline we're currently
9422 : * replaying.
9423 : *
9424 : * There is no guarantee that the WAL segments will be useful on the
9425 : * current timeline; if recovery proceeds to a new timeline right after
9426 : * this, the pre-allocated WAL segments on this timeline will not be used,
9427 : * and will go wasted until recycled on the next restartpoint. We'll live
9428 : * with that.
9429 : */
9430 30 : if (RecoveryInProgress())
9431 30 : ThisTimeLineID = replayTLI;
9432 :
9433 30 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr);
9434 :
9435 : /*
9436 : * Make more log segments if needed. (Do this after recycling old log
9437 : * segments, since that may supply some of the needed files.)
9438 : */
9439 30 : PreallocXlogFiles(endptr);
9440 :
9441 : /*
9442 : * ThisTimeLineID is normally not set when we're still in recovery.
9443 : * However, recycling/preallocating segments above needed ThisTimeLineID
9444 : * to determine which timeline to install the segments on. Reset it now,
9445 : * to restore the normal state of affairs for debugging purposes.
9446 : */
9447 30 : if (RecoveryInProgress())
9448 30 : ThisTimeLineID = 0;
9449 :
9450 : /*
9451 : * Truncate pg_subtrans if possible. We can throw away all data before
9452 : * the oldest XMIN of any running transaction. No future transaction will
9453 : * attempt to reference any pg_subtrans entry older than that (see Asserts
9454 : * in subtrans.c). When hot standby is disabled, though, we mustn't do
9455 : * this because StartupSUBTRANS hasn't been called yet.
9456 : */
9457 30 : if (EnableHotStandby)
9458 30 : TruncateSUBTRANS(GetOldestXmin(NULL, PROCARRAY_FLAGS_DEFAULT));
9459 :
9460 : /* Real work is done, but log and update before releasing lock. */
9461 30 : LogCheckpointEnd(true);
9462 :
9463 30 : xtime = GetLatestXTime();
9464 30 : ereport((log_checkpoints ? LOG : DEBUG2),
9465 : (errmsg("recovery restart point at %X/%X",
9466 : (uint32) (lastCheckPoint.redo >> 32), (uint32) lastCheckPoint.redo),
9467 : xtime ? errdetail("Last completed transaction was at log time %s.",
9468 : timestamptz_to_str(xtime)) : 0));
9469 :
9470 30 : LWLockRelease(CheckpointLock);
9471 :
9472 : /*
9473 : * Finally, execute archive_cleanup_command, if any.
9474 : */
9475 30 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
9476 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
9477 : "archive_cleanup_command",
9478 : false);
9479 :
9480 30 : return true;
9481 : }
9482 :
9483 : /*
9484 : * Report availability of WAL for the given target LSN
9485 : * (typically a slot's restart_lsn)
9486 : *
9487 : * Returns one of the following enum values:
9488 : * * WALAVAIL_NORMAL means targetLSN is available because it is in the range
9489 : * of max_wal_size.
9490 : *
9491 : * * WALAVAIL_PRESERVED means it is still available by preserving extra
9492 : * segments beyond max_wal_size. If max_slot_wal_keep_size is smaller
9493 : * than max_wal_size, this state is not returned.
9494 : *
9495 : * * WALAVAIL_REMOVED means it is definitely lost. A replication stream on
9496 : * a slot with this LSN cannot continue.
9497 : *
9498 : * * WALAVAIL_INVALID_LSN means the slot hasn't been set to reserve WAL.
9499 : */
9500 : WALAvailability
9501 182 : GetWALAvailability(XLogRecPtr targetLSN)
9502 : {
9503 : XLogRecPtr currpos; /* current write LSN */
9504 : XLogSegNo currSeg; /* segid of currpos */
9505 : XLogSegNo targetSeg; /* segid of targetLSN */
9506 : XLogSegNo oldestSeg; /* actual oldest segid */
9507 : XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
9508 182 : XLogSegNo oldestSlotSeg = InvalidXLogRecPtr; /* oldest segid kept by
9509 : * slot */
9510 : uint64 keepSegs;
9511 :
9512 : /* slot does not reserve WAL. Either deactivated, or has never been active */
9513 182 : if (XLogRecPtrIsInvalid(targetLSN))
9514 10 : return WALAVAIL_INVALID_LSN;
9515 :
9516 172 : currpos = GetXLogWriteRecPtr();
9517 :
9518 : /* calculate oldest segment currently needed by slots */
9519 172 : XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
9520 172 : KeepLogSeg(currpos, &oldestSlotSeg);
9521 :
9522 : /*
9523 : * Find the oldest extant segment file. We get 1 until checkpoint removes
9524 : * the first WAL segment file since startup, which causes the status being
9525 : * wrong under certain abnormal conditions but that doesn't actually harm.
9526 : */
9527 172 : oldestSeg = XLogGetLastRemovedSegno() + 1;
9528 :
9529 : /* calculate oldest segment by max_wal_size and wal_keep_segments */
9530 172 : XLByteToSeg(currpos, currSeg, wal_segment_size);
9531 344 : keepSegs = ConvertToXSegs(Max(max_wal_size_mb, wal_keep_segments),
9532 172 : wal_segment_size) + 1;
9533 :
9534 172 : if (currSeg > keepSegs)
9535 16 : oldestSegMaxWalSize = currSeg - keepSegs;
9536 : else
9537 156 : oldestSegMaxWalSize = 1;
9538 :
9539 : /*
9540 : * If max_slot_wal_keep_size has changed after the last call, the segment
9541 : * that would been kept by the current setting might have been lost by the
9542 : * previous setting. No point in showing normal or keeping status values
9543 : * if the targetSeg is known to be lost.
9544 : */
9545 172 : if (targetSeg >= oldestSeg)
9546 : {
9547 : /*
9548 : * show "normal" when targetSeg is within max_wal_size, even if
9549 : * max_slot_wal_keep_size is smaller than max_wal_size.
9550 : */
9551 172 : if ((max_slot_wal_keep_size_mb <= 0 ||
9552 172 : max_slot_wal_keep_size_mb >= max_wal_size_mb) &&
9553 : oldestSegMaxWalSize <= targetSeg)
9554 168 : return WALAVAIL_NORMAL;
9555 :
9556 : /* being retained by slots */
9557 4 : if (oldestSlotSeg <= targetSeg)
9558 2 : return WALAVAIL_RESERVED;
9559 : }
9560 :
9561 : /* Definitely lost */
9562 2 : return WALAVAIL_REMOVED;
9563 : }
9564 :
9565 :
9566 : /*
9567 : * Retreat *logSegNo to the last segment that we need to retain because of
9568 : * either wal_keep_segments or replication slots.
9569 : *
9570 : * This is calculated by subtracting wal_keep_segments from the given xlog
9571 : * location, recptr and by making sure that that result is below the
9572 : * requirement of replication slots. For the latter criterion we do consider
9573 : * the effects of max_slot_wal_keep_size: reserve at most that much space back
9574 : * from recptr.
9575 : */
9576 : static void
9577 3344 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
9578 : {
9579 : XLogSegNo currSegNo;
9580 : XLogSegNo segno;
9581 : XLogRecPtr keep;
9582 :
9583 3344 : XLByteToSeg(recptr, currSegNo, wal_segment_size);
9584 3344 : segno = currSegNo;
9585 :
9586 : /*
9587 : * Calculate how many segments are kept by slots first, adjusting for
9588 : * max_slot_wal_keep_size.
9589 : */
9590 3344 : keep = XLogGetReplicationSlotMinimumLSN();
9591 3344 : if (keep != InvalidXLogRecPtr)
9592 : {
9593 238 : XLByteToSeg(keep, segno, wal_segment_size);
9594 :
9595 : /* Cap by max_slot_wal_keep_size ... */
9596 238 : if (max_slot_wal_keep_size_mb >= 0)
9597 : {
9598 : XLogRecPtr slot_keep_segs;
9599 :
9600 10 : slot_keep_segs =
9601 10 : ConvertToXSegs(max_slot_wal_keep_size_mb, wal_segment_size);
9602 :
9603 10 : if (currSegNo - segno > slot_keep_segs)
9604 2 : segno = currSegNo - slot_keep_segs;
9605 : }
9606 : }
9607 :
9608 : /* but, keep at least wal_keep_segments if that's set */
9609 3344 : if (wal_keep_segments > 0 && currSegNo - segno < wal_keep_segments)
9610 : {
9611 : /* avoid underflow, don't go below 1 */
9612 90 : if (currSegNo <= wal_keep_segments)
9613 88 : segno = 1;
9614 : else
9615 2 : segno = currSegNo - wal_keep_segments;
9616 : }
9617 :
9618 : /* don't delete WAL segments newer than the calculated segment */
9619 3344 : if (XLogRecPtrIsInvalid(*logSegNo) || segno < *logSegNo)
9620 274 : *logSegNo = segno;
9621 3344 : }
9622 :
9623 : /*
9624 : * Write a NEXTOID log record
9625 : */
9626 : void
9627 1140 : XLogPutNextOid(Oid nextOid)
9628 : {
9629 1140 : XLogBeginInsert();
9630 1140 : XLogRegisterData((char *) (&nextOid), sizeof(Oid));
9631 1140 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
9632 :
9633 : /*
9634 : * We need not flush the NEXTOID record immediately, because any of the
9635 : * just-allocated OIDs could only reach disk as part of a tuple insert or
9636 : * update that would have its own XLOG record that must follow the NEXTOID
9637 : * record. Therefore, the standard buffer LSN interlock applied to those
9638 : * records will ensure no such OID reaches disk before the NEXTOID record
9639 : * does.
9640 : *
9641 : * Note, however, that the above statement only covers state "within" the
9642 : * database. When we use a generated OID as a file or directory name, we
9643 : * are in a sense violating the basic WAL rule, because that filesystem
9644 : * change may reach disk before the NEXTOID WAL record does. The impact
9645 : * of this is that if a database crash occurs immediately afterward, we
9646 : * might after restart re-generate the same OID and find that it conflicts
9647 : * with the leftover file or directory. But since for safety's sake we
9648 : * always loop until finding a nonconflicting filename, this poses no real
9649 : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
9650 : */
9651 1140 : }
9652 :
9653 : /*
9654 : * Write an XLOG SWITCH record.
9655 : *
9656 : * Here we just blindly issue an XLogInsert request for the record.
9657 : * All the magic happens inside XLogInsert.
9658 : *
9659 : * The return value is either the end+1 address of the switch record,
9660 : * or the end+1 address of the prior segment if we did not need to
9661 : * write a switch record because we are already at segment start.
9662 : */
9663 : XLogRecPtr
9664 344 : RequestXLogSwitch(bool mark_unimportant)
9665 : {
9666 : XLogRecPtr RecPtr;
9667 :
9668 : /* XLOG SWITCH has no data */
9669 344 : XLogBeginInsert();
9670 :
9671 344 : if (mark_unimportant)
9672 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
9673 344 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
9674 :
9675 344 : return RecPtr;
9676 : }
9677 :
9678 : /*
9679 : * Write a RESTORE POINT record
9680 : */
9681 : XLogRecPtr
9682 2 : XLogRestorePoint(const char *rpName)
9683 : {
9684 : XLogRecPtr RecPtr;
9685 : xl_restore_point xlrec;
9686 :
9687 2 : xlrec.rp_time = GetCurrentTimestamp();
9688 2 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
9689 :
9690 2 : XLogBeginInsert();
9691 2 : XLogRegisterData((char *) &xlrec, sizeof(xl_restore_point));
9692 :
9693 2 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
9694 :
9695 2 : ereport(LOG,
9696 : (errmsg("restore point \"%s\" created at %X/%X",
9697 : rpName, (uint32) (RecPtr >> 32), (uint32) RecPtr)));
9698 :
9699 2 : return RecPtr;
9700 : }
9701 :
9702 : /*
9703 : * Check if any of the GUC parameters that are critical for hot standby
9704 : * have changed, and update the value in pg_control file if necessary.
9705 : */
9706 : static void
9707 1356 : XLogReportParameters(void)
9708 : {
9709 1356 : if (wal_level != ControlFile->wal_level ||
9710 1210 : wal_log_hints != ControlFile->wal_log_hints ||
9711 1138 : MaxConnections != ControlFile->MaxConnections ||
9712 1136 : max_worker_processes != ControlFile->max_worker_processes ||
9713 1136 : max_wal_senders != ControlFile->max_wal_senders ||
9714 1136 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
9715 1008 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
9716 1008 : track_commit_timestamp != ControlFile->track_commit_timestamp)
9717 : {
9718 : /*
9719 : * The change in number of backend slots doesn't need to be WAL-logged
9720 : * if archiving is not enabled, as you can't start archive recovery
9721 : * with wal_level=minimal anyway. We don't really care about the
9722 : * values in pg_control either if wal_level=minimal, but seems better
9723 : * to keep them up-to-date to avoid confusion.
9724 : */
9725 360 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
9726 : {
9727 : xl_parameter_change xlrec;
9728 : XLogRecPtr recptr;
9729 :
9730 360 : xlrec.MaxConnections = MaxConnections;
9731 360 : xlrec.max_worker_processes = max_worker_processes;
9732 360 : xlrec.max_wal_senders = max_wal_senders;
9733 360 : xlrec.max_prepared_xacts = max_prepared_xacts;
9734 360 : xlrec.max_locks_per_xact = max_locks_per_xact;
9735 360 : xlrec.wal_level = wal_level;
9736 360 : xlrec.wal_log_hints = wal_log_hints;
9737 360 : xlrec.track_commit_timestamp = track_commit_timestamp;
9738 :
9739 360 : XLogBeginInsert();
9740 360 : XLogRegisterData((char *) &xlrec, sizeof(xlrec));
9741 :
9742 360 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
9743 360 : XLogFlush(recptr);
9744 : }
9745 :
9746 360 : ControlFile->MaxConnections = MaxConnections;
9747 360 : ControlFile->max_worker_processes = max_worker_processes;
9748 360 : ControlFile->max_wal_senders = max_wal_senders;
9749 360 : ControlFile->max_prepared_xacts = max_prepared_xacts;
9750 360 : ControlFile->max_locks_per_xact = max_locks_per_xact;
9751 360 : ControlFile->wal_level = wal_level;
9752 360 : ControlFile->wal_log_hints = wal_log_hints;
9753 360 : ControlFile->track_commit_timestamp = track_commit_timestamp;
9754 360 : UpdateControlFile();
9755 : }
9756 1356 : }
9757 :
9758 : /*
9759 : * Update full_page_writes in shared memory, and write an
9760 : * XLOG_FPW_CHANGE record if necessary.
9761 : *
9762 : * Note: this function assumes there is no other process running
9763 : * concurrently that could update it.
9764 : */
9765 : void
9766 1778 : UpdateFullPageWrites(void)
9767 : {
9768 1778 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9769 : bool recoveryInProgress;
9770 :
9771 : /*
9772 : * Do nothing if full_page_writes has not been changed.
9773 : *
9774 : * It's safe to check the shared full_page_writes without the lock,
9775 : * because we assume that there is no concurrently running process which
9776 : * can update it.
9777 : */
9778 1778 : if (fullPageWrites == Insert->fullPageWrites)
9779 1722 : return;
9780 :
9781 : /*
9782 : * Perform this outside critical section so that the WAL insert
9783 : * initialization done by RecoveryInProgress() doesn't trigger an
9784 : * assertion failure.
9785 : */
9786 56 : recoveryInProgress = RecoveryInProgress();
9787 :
9788 56 : START_CRIT_SECTION();
9789 :
9790 : /*
9791 : * It's always safe to take full page images, even when not strictly
9792 : * required, but not the other round. So if we're setting full_page_writes
9793 : * to true, first set it true and then write the WAL record. If we're
9794 : * setting it to false, first write the WAL record and then set the global
9795 : * flag.
9796 : */
9797 56 : if (fullPageWrites)
9798 : {
9799 54 : WALInsertLockAcquireExclusive();
9800 54 : Insert->fullPageWrites = true;
9801 54 : WALInsertLockRelease();
9802 : }
9803 :
9804 : /*
9805 : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
9806 : * full_page_writes during archive recovery, if required.
9807 : */
9808 56 : if (XLogStandbyInfoActive() && !recoveryInProgress)
9809 : {
9810 0 : XLogBeginInsert();
9811 0 : XLogRegisterData((char *) (&fullPageWrites), sizeof(bool));
9812 :
9813 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
9814 : }
9815 :
9816 56 : if (!fullPageWrites)
9817 : {
9818 2 : WALInsertLockAcquireExclusive();
9819 2 : Insert->fullPageWrites = false;
9820 2 : WALInsertLockRelease();
9821 : }
9822 56 : END_CRIT_SECTION();
9823 : }
9824 :
9825 : /*
9826 : * Check that it's OK to switch to new timeline during recovery.
9827 : *
9828 : * 'lsn' is the address of the shutdown checkpoint record we're about to
9829 : * replay. (Currently, timeline can only change at a shutdown checkpoint).
9830 : */
9831 : static void
9832 10 : checkTimeLineSwitch(XLogRecPtr lsn, TimeLineID newTLI, TimeLineID prevTLI)
9833 : {
9834 : /* Check that the record agrees on what the current (old) timeline is */
9835 10 : if (prevTLI != ThisTimeLineID)
9836 0 : ereport(PANIC,
9837 : (errmsg("unexpected previous timeline ID %u (current timeline ID %u) in checkpoint record",
9838 : prevTLI, ThisTimeLineID)));
9839 :
9840 : /*
9841 : * The new timeline better be in the list of timelines we expect to see,
9842 : * according to the timeline history. It should also not decrease.
9843 : */
9844 10 : if (newTLI < ThisTimeLineID || !tliInHistory(newTLI, expectedTLEs))
9845 0 : ereport(PANIC,
9846 : (errmsg("unexpected timeline ID %u (after %u) in checkpoint record",
9847 : newTLI, ThisTimeLineID)));
9848 :
9849 : /*
9850 : * If we have not yet reached min recovery point, and we're about to
9851 : * switch to a timeline greater than the timeline of the min recovery
9852 : * point: trouble. After switching to the new timeline, we could not
9853 : * possibly visit the min recovery point on the correct timeline anymore.
9854 : * This can happen if there is a newer timeline in the archive that
9855 : * branched before the timeline the min recovery point is on, and you
9856 : * attempt to do PITR to the new timeline.
9857 : */
9858 10 : if (!XLogRecPtrIsInvalid(minRecoveryPoint) &&
9859 8 : lsn < minRecoveryPoint &&
9860 0 : newTLI > minRecoveryPointTLI)
9861 0 : ereport(PANIC,
9862 : (errmsg("unexpected timeline ID %u in checkpoint record, before reaching minimum recovery point %X/%X on timeline %u",
9863 : newTLI,
9864 : (uint32) (minRecoveryPoint >> 32),
9865 : (uint32) minRecoveryPoint,
9866 : minRecoveryPointTLI)));
9867 :
9868 : /* Looks good */
9869 10 : }
9870 :
9871 : /*
9872 : * XLOG resource manager's routines
9873 : *
9874 : * Definitions of info values are in include/catalog/pg_control.h, though
9875 : * not all record types are related to control file updates.
9876 : */
9877 : void
9878 3040 : xlog_redo(XLogReaderState *record)
9879 : {
9880 3040 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9881 3040 : XLogRecPtr lsn = record->EndRecPtr;
9882 :
9883 : /* in XLOG rmgr, backup blocks are only used by XLOG_FPI records */
9884 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
9885 : !XLogRecHasAnyBlockRefs(record));
9886 :
9887 3040 : if (info == XLOG_NEXTOID)
9888 : {
9889 : Oid nextOid;
9890 :
9891 : /*
9892 : * We used to try to take the maximum of ShmemVariableCache->nextOid
9893 : * and the recorded nextOid, but that fails if the OID counter wraps
9894 : * around. Since no OID allocation should be happening during replay
9895 : * anyway, better to just believe the record exactly. We still take
9896 : * OidGenLock while setting the variable, just in case.
9897 : */
9898 100 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
9899 100 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9900 100 : ShmemVariableCache->nextOid = nextOid;
9901 100 : ShmemVariableCache->oidCount = 0;
9902 100 : LWLockRelease(OidGenLock);
9903 : }
9904 2940 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
9905 : {
9906 : CheckPoint checkPoint;
9907 :
9908 30 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
9909 : /* In a SHUTDOWN checkpoint, believe the counters exactly */
9910 30 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
9911 30 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
9912 30 : LWLockRelease(XidGenLock);
9913 30 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
9914 30 : ShmemVariableCache->nextOid = checkPoint.nextOid;
9915 30 : ShmemVariableCache->oidCount = 0;
9916 30 : LWLockRelease(OidGenLock);
9917 30 : MultiXactSetNextMXact(checkPoint.nextMulti,
9918 : checkPoint.nextMultiOffset);
9919 :
9920 30 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
9921 : checkPoint.oldestMultiDB);
9922 :
9923 : /*
9924 : * No need to set oldestClogXid here as well; it'll be set when we
9925 : * redo an xl_clog_truncate if it changed since initialization.
9926 : */
9927 30 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
9928 :
9929 : /*
9930 : * If we see a shutdown checkpoint while waiting for an end-of-backup
9931 : * record, the backup was canceled and the end-of-backup record will
9932 : * never arrive.
9933 : */
9934 30 : if (ArchiveRecoveryRequested &&
9935 30 : !XLogRecPtrIsInvalid(ControlFile->backupStartPoint) &&
9936 0 : XLogRecPtrIsInvalid(ControlFile->backupEndPoint))
9937 0 : ereport(PANIC,
9938 : (errmsg("online backup was canceled, recovery cannot continue")));
9939 :
9940 : /*
9941 : * If we see a shutdown checkpoint, we know that nothing was running
9942 : * on the master at this point. So fake-up an empty running-xacts
9943 : * record and use that here and now. Recover additional standby state
9944 : * for prepared transactions.
9945 : */
9946 30 : if (standbyState >= STANDBY_INITIALIZED)
9947 : {
9948 : TransactionId *xids;
9949 : int nxids;
9950 : TransactionId oldestActiveXID;
9951 : TransactionId latestCompletedXid;
9952 : RunningTransactionsData running;
9953 :
9954 30 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
9955 :
9956 : /*
9957 : * Construct a RunningTransactions snapshot representing a shut
9958 : * down server, with only prepared transactions still alive. We're
9959 : * never overflowed at this point because all subxids are listed
9960 : * with their parent prepared transactions.
9961 : */
9962 30 : running.xcnt = nxids;
9963 30 : running.subxcnt = 0;
9964 30 : running.subxid_overflow = false;
9965 30 : running.nextXid = XidFromFullTransactionId(checkPoint.nextFullXid);
9966 30 : running.oldestRunningXid = oldestActiveXID;
9967 30 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextFullXid);
9968 30 : TransactionIdRetreat(latestCompletedXid);
9969 : Assert(TransactionIdIsNormal(latestCompletedXid));
9970 30 : running.latestCompletedXid = latestCompletedXid;
9971 30 : running.xids = xids;
9972 :
9973 30 : ProcArrayApplyRecoveryInfo(&running);
9974 :
9975 30 : StandbyRecoverPreparedTransactions();
9976 : }
9977 :
9978 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
9979 30 : ControlFile->checkPointCopy.nextFullXid = checkPoint.nextFullXid;
9980 :
9981 : /* Update shared-memory copy of checkpoint XID/epoch */
9982 30 : SpinLockAcquire(&XLogCtl->info_lck);
9983 30 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
9984 30 : SpinLockRelease(&XLogCtl->info_lck);
9985 :
9986 : /*
9987 : * We should've already switched to the new TLI before replaying this
9988 : * record.
9989 : */
9990 30 : if (checkPoint.ThisTimeLineID != ThisTimeLineID)
9991 0 : ereport(PANIC,
9992 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
9993 : checkPoint.ThisTimeLineID, ThisTimeLineID)));
9994 :
9995 30 : RecoveryRestartPoint(&checkPoint);
9996 : }
9997 2910 : else if (info == XLOG_CHECKPOINT_ONLINE)
9998 : {
9999 : CheckPoint checkPoint;
10000 :
10001 172 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
10002 : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
10003 172 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
10004 172 : if (FullTransactionIdPrecedes(ShmemVariableCache->nextFullXid,
10005 : checkPoint.nextFullXid))
10006 0 : ShmemVariableCache->nextFullXid = checkPoint.nextFullXid;
10007 172 : LWLockRelease(XidGenLock);
10008 :
10009 : /*
10010 : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
10011 : * to track OID assignment through XLOG_NEXTOID records. The nextOid
10012 : * counter is from the start of the checkpoint and might well be stale
10013 : * compared to later XLOG_NEXTOID records. We could try to take the
10014 : * maximum of the nextOid counter and our latest value, but since
10015 : * there's no particular guarantee about the speed with which the OID
10016 : * counter wraps around, that's a risky thing to do. In any case,
10017 : * users of the nextOid counter are required to avoid assignment of
10018 : * duplicates, so that a somewhat out-of-date value should be safe.
10019 : */
10020 :
10021 : /* Handle multixact */
10022 172 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
10023 : checkPoint.nextMultiOffset);
10024 :
10025 : /*
10026 : * NB: This may perform multixact truncation when replaying WAL
10027 : * generated by an older primary.
10028 : */
10029 172 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
10030 : checkPoint.oldestMultiDB);
10031 172 : if (TransactionIdPrecedes(ShmemVariableCache->oldestXid,
10032 : checkPoint.oldestXid))
10033 0 : SetTransactionIdLimit(checkPoint.oldestXid,
10034 : checkPoint.oldestXidDB);
10035 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
10036 172 : ControlFile->checkPointCopy.nextFullXid = checkPoint.nextFullXid;
10037 :
10038 : /* Update shared-memory copy of checkpoint XID/epoch */
10039 172 : SpinLockAcquire(&XLogCtl->info_lck);
10040 172 : XLogCtl->ckptFullXid = checkPoint.nextFullXid;
10041 172 : SpinLockRelease(&XLogCtl->info_lck);
10042 :
10043 : /* TLI should not change in an on-line checkpoint */
10044 172 : if (checkPoint.ThisTimeLineID != ThisTimeLineID)
10045 0 : ereport(PANIC,
10046 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
10047 : checkPoint.ThisTimeLineID, ThisTimeLineID)));
10048 :
10049 172 : RecoveryRestartPoint(&checkPoint);
10050 : }
10051 2738 : else if (info == XLOG_END_OF_RECOVERY)
10052 : {
10053 : xl_end_of_recovery xlrec;
10054 :
10055 10 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
10056 :
10057 : /*
10058 : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
10059 : * but this case is rarer and harder to test, so the benefit doesn't
10060 : * outweigh the potential extra cost of maintenance.
10061 : */
10062 :
10063 : /*
10064 : * We should've already switched to the new TLI before replaying this
10065 : * record.
10066 : */
10067 10 : if (xlrec.ThisTimeLineID != ThisTimeLineID)
10068 0 : ereport(PANIC,
10069 : (errmsg("unexpected timeline ID %u (should be %u) in checkpoint record",
10070 : xlrec.ThisTimeLineID, ThisTimeLineID)));
10071 : }
10072 2728 : else if (info == XLOG_NOOP)
10073 : {
10074 : /* nothing to do here */
10075 : }
10076 2728 : else if (info == XLOG_SWITCH)
10077 : {
10078 : /* nothing to do here */
10079 : }
10080 2612 : else if (info == XLOG_RESTORE_POINT)
10081 : {
10082 : /* nothing to do here */
10083 : }
10084 2610 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
10085 : {
10086 : /*
10087 : * Full-page image (FPI) records contain nothing else but a backup
10088 : * block (or multiple backup blocks). Every block reference must
10089 : * include a full-page image - otherwise there would be no point in
10090 : * this record.
10091 : *
10092 : * No recovery conflicts are generated by these generic records - if a
10093 : * resource manager needs to generate conflicts, it has to define a
10094 : * separate WAL record type and redo routine.
10095 : *
10096 : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
10097 : * WAL- logged because of a hint bit update. They are only generated
10098 : * when checksums are enabled. There is no difference in handling
10099 : * XLOG_FPI and XLOG_FPI_FOR_HINT records, they use a different info
10100 : * code just to distinguish them for statistics purposes.
10101 : */
10102 5334 : for (uint8 block_id = 0; block_id <= record->max_block_id; block_id++)
10103 : {
10104 : Buffer buffer;
10105 :
10106 2810 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
10107 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
10108 2810 : UnlockReleaseBuffer(buffer);
10109 : }
10110 : }
10111 86 : else if (info == XLOG_BACKUP_END)
10112 : {
10113 : XLogRecPtr startpoint;
10114 :
10115 68 : memcpy(&startpoint, XLogRecGetData(record), sizeof(startpoint));
10116 :
10117 68 : if (ControlFile->backupStartPoint == startpoint)
10118 : {
10119 : /*
10120 : * We have reached the end of base backup, the point where
10121 : * pg_stop_backup() was done. The data on disk is now consistent.
10122 : * Reset backupStartPoint, and update minRecoveryPoint to make
10123 : * sure we don't allow starting up at an earlier point even if
10124 : * recovery is stopped and restarted soon after this.
10125 : */
10126 58 : elog(DEBUG1, "end of backup reached");
10127 :
10128 58 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
10129 :
10130 58 : if (ControlFile->minRecoveryPoint < lsn)
10131 : {
10132 58 : ControlFile->minRecoveryPoint = lsn;
10133 58 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
10134 : }
10135 58 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
10136 58 : ControlFile->backupEndRequired = false;
10137 58 : UpdateControlFile();
10138 :
10139 58 : LWLockRelease(ControlFileLock);
10140 : }
10141 : }
10142 18 : else if (info == XLOG_PARAMETER_CHANGE)
10143 : {
10144 : xl_parameter_change xlrec;
10145 :
10146 : /* Update our copy of the parameters in pg_control */
10147 18 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
10148 :
10149 18 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
10150 18 : ControlFile->MaxConnections = xlrec.MaxConnections;
10151 18 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
10152 18 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
10153 18 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
10154 18 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
10155 18 : ControlFile->wal_level = xlrec.wal_level;
10156 18 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
10157 :
10158 : /*
10159 : * Update minRecoveryPoint to ensure that if recovery is aborted, we
10160 : * recover back up to this point before allowing hot standby again.
10161 : * This is important if the max_* settings are decreased, to ensure
10162 : * you don't run queries against the WAL preceding the change. The
10163 : * local copies cannot be updated as long as crash recovery is
10164 : * happening and we expect all the WAL to be replayed.
10165 : */
10166 18 : if (InArchiveRecovery)
10167 : {
10168 4 : minRecoveryPoint = ControlFile->minRecoveryPoint;
10169 4 : minRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
10170 : }
10171 18 : if (minRecoveryPoint != InvalidXLogRecPtr && minRecoveryPoint < lsn)
10172 : {
10173 4 : ControlFile->minRecoveryPoint = lsn;
10174 4 : ControlFile->minRecoveryPointTLI = ThisTimeLineID;
10175 : }
10176 :
10177 18 : CommitTsParameterChange(xlrec.track_commit_timestamp,
10178 18 : ControlFile->track_commit_timestamp);
10179 18 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
10180 :
10181 18 : UpdateControlFile();
10182 18 : LWLockRelease(ControlFileLock);
10183 :
10184 : /* Check to see if any parameter change gives a problem on recovery */
10185 18 : CheckRequiredParameterValues();
10186 : }
10187 0 : else if (info == XLOG_FPW_CHANGE)
10188 : {
10189 : bool fpw;
10190 :
10191 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
10192 :
10193 : /*
10194 : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
10195 : * do_pg_start_backup() and do_pg_stop_backup() can check whether
10196 : * full_page_writes has been disabled during online backup.
10197 : */
10198 0 : if (!fpw)
10199 : {
10200 0 : SpinLockAcquire(&XLogCtl->info_lck);
10201 0 : if (XLogCtl->lastFpwDisableRecPtr < ReadRecPtr)
10202 0 : XLogCtl->lastFpwDisableRecPtr = ReadRecPtr;
10203 0 : SpinLockRelease(&XLogCtl->info_lck);
10204 : }
10205 :
10206 : /* Keep track of full_page_writes */
10207 0 : lastFullPageWrites = fpw;
10208 : }
10209 3040 : }
10210 :
10211 : #ifdef WAL_DEBUG
10212 :
10213 : static void
10214 : xlog_outrec(StringInfo buf, XLogReaderState *record)
10215 : {
10216 : int block_id;
10217 :
10218 : appendStringInfo(buf, "prev %X/%X; xid %u",
10219 : (uint32) (XLogRecGetPrev(record) >> 32),
10220 : (uint32) XLogRecGetPrev(record),
10221 : XLogRecGetXid(record));
10222 :
10223 : appendStringInfo(buf, "; len %u",
10224 : XLogRecGetDataLen(record));
10225 :
10226 : /* decode block references */
10227 : for (block_id = 0; block_id <= record->max_block_id; block_id++)
10228 : {
10229 : RelFileNode rnode;
10230 : ForkNumber forknum;
10231 : BlockNumber blk;
10232 :
10233 : if (!XLogRecHasBlockRef(record, block_id))
10234 : continue;
10235 :
10236 : XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blk);
10237 : if (forknum != MAIN_FORKNUM)
10238 : appendStringInfo(buf, "; blkref #%u: rel %u/%u/%u, fork %u, blk %u",
10239 : block_id,
10240 : rnode.spcNode, rnode.dbNode, rnode.relNode,
10241 : forknum,
10242 : blk);
10243 : else
10244 : appendStringInfo(buf, "; blkref #%u: rel %u/%u/%u, blk %u",
10245 : block_id,
10246 : rnode.spcNode, rnode.dbNode, rnode.relNode,
10247 : blk);
10248 : if (XLogRecHasBlockImage(record, block_id))
10249 : appendStringInfoString(buf, " FPW");
10250 : }
10251 : }
10252 : #endif /* WAL_DEBUG */
10253 :
10254 : /*
10255 : * Returns a string describing an XLogRecord, consisting of its identity
10256 : * optionally followed by a colon, a space, and a further description.
10257 : */
10258 : static void
10259 6 : xlog_outdesc(StringInfo buf, XLogReaderState *record)
10260 : {
10261 6 : RmgrId rmid = XLogRecGetRmid(record);
10262 6 : uint8 info = XLogRecGetInfo(record);
10263 : const char *id;
10264 :
10265 6 : appendStringInfoString(buf, RmgrTable[rmid].rm_name);
10266 6 : appendStringInfoChar(buf, '/');
10267 :
10268 6 : id = RmgrTable[rmid].rm_identify(info);
10269 6 : if (id == NULL)
10270 0 : appendStringInfo(buf, "UNKNOWN (%X): ", info & ~XLR_INFO_MASK);
10271 : else
10272 6 : appendStringInfo(buf, "%s: ", id);
10273 :
10274 6 : RmgrTable[rmid].rm_desc(buf, record);
10275 6 : }
10276 :
10277 :
10278 : /*
10279 : * Return the (possible) sync flag used for opening a file, depending on the
10280 : * value of the GUC wal_sync_method.
10281 : */
10282 : static int
10283 9292 : get_sync_bit(int method)
10284 : {
10285 9292 : int o_direct_flag = 0;
10286 :
10287 : /* If fsync is disabled, never open in sync mode */
10288 9292 : if (!enableFsync)
10289 9260 : return 0;
10290 :
10291 : /*
10292 : * Optimize writes by bypassing kernel cache with O_DIRECT when using
10293 : * O_SYNC/O_FSYNC and O_DSYNC. But only if archiving and streaming are
10294 : * disabled, otherwise the archive command or walsender process will read
10295 : * the WAL soon after writing it, which is guaranteed to cause a physical
10296 : * read if we bypassed the kernel cache. We also skip the
10297 : * posix_fadvise(POSIX_FADV_DONTNEED) call in XLogFileClose() for the same
10298 : * reason.
10299 : *
10300 : * Never use O_DIRECT in walreceiver process for similar reasons; the WAL
10301 : * written by walreceiver is normally read by the startup process soon
10302 : * after its written. Also, walreceiver performs unaligned writes, which
10303 : * don't work with O_DIRECT, so it is required for correctness too.
10304 : */
10305 32 : if (!XLogIsNeeded() && !AmWalReceiverProcess())
10306 0 : o_direct_flag = PG_O_DIRECT;
10307 :
10308 32 : switch (method)
10309 : {
10310 : /*
10311 : * enum values for all sync options are defined even if they are
10312 : * not supported on the current platform. But if not, they are
10313 : * not included in the enum option array, and therefore will never
10314 : * be seen here.
10315 : */
10316 32 : case SYNC_METHOD_FSYNC:
10317 : case SYNC_METHOD_FSYNC_WRITETHROUGH:
10318 : case SYNC_METHOD_FDATASYNC:
10319 32 : return 0;
10320 : #ifdef OPEN_SYNC_FLAG
10321 0 : case SYNC_METHOD_OPEN:
10322 0 : return OPEN_SYNC_FLAG | o_direct_flag;
10323 : #endif
10324 : #ifdef OPEN_DATASYNC_FLAG
10325 0 : case SYNC_METHOD_OPEN_DSYNC:
10326 0 : return OPEN_DATASYNC_FLAG | o_direct_flag;
10327 : #endif
10328 0 : default:
10329 : /* can't happen (unless we are out of sync with option array) */
10330 0 : elog(ERROR, "unrecognized wal_sync_method: %d", method);
10331 : return 0; /* silence warning */
10332 : }
10333 : }
10334 :
10335 : /*
10336 : * GUC support
10337 : */
10338 : void
10339 2184 : assign_xlog_sync_method(int new_sync_method, void *extra)
10340 : {
10341 2184 : if (sync_method != new_sync_method)
10342 : {
10343 : /*
10344 : * To ensure that no blocks escape unsynced, force an fsync on the
10345 : * currently open log segment (if any). Also, if the open flag is
10346 : * changing, close the log file so it will be reopened (with new flag
10347 : * bit) at next use.
10348 : */
10349 0 : if (openLogFile >= 0)
10350 : {
10351 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
10352 0 : if (pg_fsync(openLogFile) != 0)
10353 : {
10354 : char xlogfname[MAXFNAMELEN];
10355 : int save_errno;
10356 :
10357 0 : save_errno = errno;
10358 0 : XLogFileName(xlogfname, ThisTimeLineID, openLogSegNo,
10359 : wal_segment_size);
10360 0 : errno = save_errno;
10361 0 : ereport(PANIC,
10362 : (errcode_for_file_access(),
10363 : errmsg("could not fsync file \"%s\": %m", xlogfname)));
10364 : }
10365 :
10366 0 : pgstat_report_wait_end();
10367 0 : if (get_sync_bit(sync_method) != get_sync_bit(new_sync_method))
10368 0 : XLogFileClose();
10369 : }
10370 : }
10371 2184 : }
10372 :
10373 :
10374 : /*
10375 : * Issue appropriate kind of fsync (if any) for an XLOG output file.
10376 : *
10377 : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
10378 : * 'segno' is for error reporting purposes.
10379 : */
10380 : void
10381 265338 : issue_xlog_fsync(int fd, XLogSegNo segno)
10382 : {
10383 265338 : char *msg = NULL;
10384 :
10385 265338 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
10386 265338 : switch (sync_method)
10387 : {
10388 0 : case SYNC_METHOD_FSYNC:
10389 0 : if (pg_fsync_no_writethrough(fd) != 0)
10390 0 : msg = _("could not fsync file \"%s\": %m");
10391 0 : break;
10392 : #ifdef HAVE_FSYNC_WRITETHROUGH
10393 : case SYNC_METHOD_FSYNC_WRITETHROUGH:
10394 : if (pg_fsync_writethrough(fd) != 0)
10395 : msg = _("could not fsync write-through file \"%s\": %m");
10396 : break;
10397 : #endif
10398 : #ifdef HAVE_FDATASYNC
10399 265338 : case SYNC_METHOD_FDATASYNC:
10400 265338 : if (pg_fdatasync(fd) != 0)
10401 0 : msg = _("could not fdatasync file \"%s\": %m");
10402 265338 : break;
10403 : #endif
10404 0 : case SYNC_METHOD_OPEN:
10405 : case SYNC_METHOD_OPEN_DSYNC:
10406 : /* write synced it already */
10407 0 : break;
10408 0 : default:
10409 0 : elog(PANIC, "unrecognized wal_sync_method: %d", sync_method);
10410 : break;
10411 : }
10412 :
10413 : /* PANIC if failed to fsync */
10414 265338 : if (msg)
10415 : {
10416 : char xlogfname[MAXFNAMELEN];
10417 0 : int save_errno = errno;
10418 :
10419 0 : XLogFileName(xlogfname, ThisTimeLineID, segno,
10420 : wal_segment_size);
10421 0 : errno = save_errno;
10422 0 : ereport(PANIC,
10423 : (errcode_for_file_access(),
10424 : errmsg(msg, xlogfname)));
10425 : }
10426 :
10427 265338 : pgstat_report_wait_end();
10428 265338 : }
10429 :
10430 : /*
10431 : * do_pg_start_backup
10432 : *
10433 : * Utility function called at the start of an online backup. It creates the
10434 : * necessary starting checkpoint and constructs the backup label file.
10435 : *
10436 : * There are two kind of backups: exclusive and non-exclusive. An exclusive
10437 : * backup is started with pg_start_backup(), and there can be only one active
10438 : * at a time. The backup and tablespace map files of an exclusive backup are
10439 : * written to $PGDATA/backup_label and $PGDATA/tablespace_map, and they are
10440 : * removed by pg_stop_backup().
10441 : *
10442 : * A non-exclusive backup is used for the streaming base backups (see
10443 : * src/backend/replication/basebackup.c). The difference to exclusive backups
10444 : * is that the backup label and tablespace map files are not written to disk.
10445 : * Instead, their would-be contents are returned in *labelfile and *tblspcmapfile,
10446 : * and the caller is responsible for including them in the backup archive as
10447 : * 'backup_label' and 'tablespace_map'. There can be many non-exclusive backups
10448 : * active at the same time, and they don't conflict with an exclusive backup
10449 : * either.
10450 : *
10451 : * tablespaces is required only when this function is called while
10452 : * the streaming base backup requested by pg_basebackup is running.
10453 : * NULL should be specified otherwise.
10454 : *
10455 : * tblspcmapfile is required mainly for tar format in windows as native windows
10456 : * utilities are not able to create symlinks while extracting files from tar.
10457 : * However for consistency, the same is used for all platforms.
10458 : *
10459 : * needtblspcmapfile is true for the cases (exclusive backup and for
10460 : * non-exclusive backup only when tar format is used for taking backup)
10461 : * when backup needs to generate tablespace_map file, it is used to
10462 : * embed escape character before newline character in tablespace path.
10463 : *
10464 : * Returns the minimum WAL location that must be present to restore from this
10465 : * backup, and the corresponding timeline ID in *starttli_p.
10466 : *
10467 : * Every successfully started non-exclusive backup must be stopped by calling
10468 : * do_pg_stop_backup() or do_pg_abort_backup().
10469 : *
10470 : * It is the responsibility of the caller of this function to verify the
10471 : * permissions of the calling user!
10472 : */
10473 : XLogRecPtr
10474 146 : do_pg_start_backup(const char *backupidstr, bool fast, TimeLineID *starttli_p,
10475 : StringInfo labelfile, List **tablespaces,
10476 : StringInfo tblspcmapfile, bool infotbssize,
10477 : bool needtblspcmapfile)
10478 : {
10479 146 : bool exclusive = (labelfile == NULL);
10480 146 : bool backup_started_in_recovery = false;
10481 : XLogRecPtr checkpointloc;
10482 : XLogRecPtr startpoint;
10483 : TimeLineID starttli;
10484 : pg_time_t stamp_time;
10485 : char strfbuf[128];
10486 : char xlogfilename[MAXFNAMELEN];
10487 : XLogSegNo _logSegNo;
10488 : struct stat stat_buf;
10489 : FILE *fp;
10490 :
10491 146 : backup_started_in_recovery = RecoveryInProgress();
10492 :
10493 : /*
10494 : * Currently only non-exclusive backup can be taken during recovery.
10495 : */
10496 146 : if (backup_started_in_recovery && exclusive)
10497 0 : ereport(ERROR,
10498 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10499 : errmsg("recovery is in progress"),
10500 : errhint("WAL control functions cannot be executed during recovery.")));
10501 :
10502 : /*
10503 : * During recovery, we don't need to check WAL level. Because, if WAL
10504 : * level is not sufficient, it's impossible to get here during recovery.
10505 : */
10506 146 : if (!backup_started_in_recovery && !XLogIsNeeded())
10507 0 : ereport(ERROR,
10508 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10509 : errmsg("WAL level not sufficient for making an online backup"),
10510 : errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
10511 :
10512 146 : if (strlen(backupidstr) > MAXPGPATH)
10513 0 : ereport(ERROR,
10514 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
10515 : errmsg("backup label too long (max %d bytes)",
10516 : MAXPGPATH)));
10517 :
10518 : /*
10519 : * Mark backup active in shared memory. We must do full-page WAL writes
10520 : * during an on-line backup even if not doing so at other times, because
10521 : * it's quite possible for the backup dump to obtain a "torn" (partially
10522 : * written) copy of a database page if it reads the page concurrently with
10523 : * our write to the same page. This can be fixed as long as the first
10524 : * write to the page in the WAL sequence is a full-page write. Hence, we
10525 : * turn on forcePageWrites and then force a CHECKPOINT, to ensure there
10526 : * are no dirty pages in shared memory that might get dumped while the
10527 : * backup is in progress without having a corresponding WAL record. (Once
10528 : * the backup is complete, we need not force full-page writes anymore,
10529 : * since we expect that any pages not modified during the backup interval
10530 : * must have been correctly captured by the backup.)
10531 : *
10532 : * Note that forcePageWrites has no effect during an online backup from
10533 : * the standby.
10534 : *
10535 : * We must hold all the insertion locks to change the value of
10536 : * forcePageWrites, to ensure adequate interlocking against
10537 : * XLogInsertRecord().
10538 : */
10539 146 : WALInsertLockAcquireExclusive();
10540 146 : if (exclusive)
10541 : {
10542 : /*
10543 : * At first, mark that we're now starting an exclusive backup, to
10544 : * ensure that there are no other sessions currently running
10545 : * pg_start_backup() or pg_stop_backup().
10546 : */
10547 2 : if (XLogCtl->Insert.exclusiveBackupState != EXCLUSIVE_BACKUP_NONE)
10548 : {
10549 0 : WALInsertLockRelease();
10550 0 : ereport(ERROR,
10551 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10552 : errmsg("a backup is already in progress"),
10553 : errhint("Run pg_stop_backup() and try again.")));
10554 : }
10555 2 : XLogCtl->Insert.exclusiveBackupState = EXCLUSIVE_BACKUP_STARTING;
10556 : }
10557 : else
10558 144 : XLogCtl->Insert.nonExclusiveBackups++;
10559 146 : XLogCtl->Insert.forcePageWrites = true;
10560 146 : WALInsertLockRelease();
10561 :
10562 : /* Ensure we release forcePageWrites if fail below */
10563 146 : PG_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum) BoolGetDatum(exclusive));
10564 : {
10565 146 : bool gotUniqueStartpoint = false;
10566 : DIR *tblspcdir;
10567 : struct dirent *de;
10568 : tablespaceinfo *ti;
10569 : int datadirpathlen;
10570 :
10571 : /*
10572 : * Force an XLOG file switch before the checkpoint, to ensure that the
10573 : * WAL segment the checkpoint is written to doesn't contain pages with
10574 : * old timeline IDs. That would otherwise happen if you called
10575 : * pg_start_backup() right after restoring from a PITR archive: the
10576 : * first WAL segment containing the startup checkpoint has pages in
10577 : * the beginning with the old timeline ID. That can cause trouble at
10578 : * recovery: we won't have a history file covering the old timeline if
10579 : * pg_wal directory was not included in the base backup and the WAL
10580 : * archive was cleared too before starting the backup.
10581 : *
10582 : * This also ensures that we have emitted a WAL page header that has
10583 : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
10584 : * Therefore, if a WAL archiver (such as pglesslog) is trying to
10585 : * compress out removable backup blocks, it won't remove any that
10586 : * occur after this point.
10587 : *
10588 : * During recovery, we skip forcing XLOG file switch, which means that
10589 : * the backup taken during recovery is not available for the special
10590 : * recovery case described above.
10591 : */
10592 146 : if (!backup_started_in_recovery)
10593 142 : RequestXLogSwitch(false);
10594 :
10595 : do
10596 : {
10597 : bool checkpointfpw;
10598 :
10599 : /*
10600 : * Force a CHECKPOINT. Aside from being necessary to prevent torn
10601 : * page problems, this guarantees that two successive backup runs
10602 : * will have different checkpoint positions and hence different
10603 : * history file names, even if nothing happened in between.
10604 : *
10605 : * During recovery, establish a restartpoint if possible. We use
10606 : * the last restartpoint as the backup starting checkpoint. This
10607 : * means that two successive backup runs can have same checkpoint
10608 : * positions.
10609 : *
10610 : * Since the fact that we are executing do_pg_start_backup()
10611 : * during recovery means that checkpointer is running, we can use
10612 : * RequestCheckpoint() to establish a restartpoint.
10613 : *
10614 : * We use CHECKPOINT_IMMEDIATE only if requested by user (via
10615 : * passing fast = true). Otherwise this can take awhile.
10616 : */
10617 146 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
10618 : (fast ? CHECKPOINT_IMMEDIATE : 0));
10619 :
10620 : /*
10621 : * Now we need to fetch the checkpoint record location, and also
10622 : * its REDO pointer. The oldest point in WAL that would be needed
10623 : * to restore starting from the checkpoint is precisely the REDO
10624 : * pointer.
10625 : */
10626 146 : LWLockAcquire(ControlFileLock, LW_SHARED);
10627 146 : checkpointloc = ControlFile->checkPoint;
10628 146 : startpoint = ControlFile->checkPointCopy.redo;
10629 146 : starttli = ControlFile->checkPointCopy.ThisTimeLineID;
10630 146 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
10631 146 : LWLockRelease(ControlFileLock);
10632 :
10633 146 : if (backup_started_in_recovery)
10634 : {
10635 : XLogRecPtr recptr;
10636 :
10637 : /*
10638 : * Check to see if all WAL replayed during online backup
10639 : * (i.e., since last restartpoint used as backup starting
10640 : * checkpoint) contain full-page writes.
10641 : */
10642 4 : SpinLockAcquire(&XLogCtl->info_lck);
10643 4 : recptr = XLogCtl->lastFpwDisableRecPtr;
10644 4 : SpinLockRelease(&XLogCtl->info_lck);
10645 :
10646 4 : if (!checkpointfpw || startpoint <= recptr)
10647 0 : ereport(ERROR,
10648 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10649 : errmsg("WAL generated with full_page_writes=off was replayed "
10650 : "since last restartpoint"),
10651 : errhint("This means that the backup being taken on the standby "
10652 : "is corrupt and should not be used. "
10653 : "Enable full_page_writes and run CHECKPOINT on the master, "
10654 : "and then try an online backup again.")));
10655 :
10656 : /*
10657 : * During recovery, since we don't use the end-of-backup WAL
10658 : * record and don't write the backup history file, the
10659 : * starting WAL location doesn't need to be unique. This means
10660 : * that two base backups started at the same time might use
10661 : * the same checkpoint as starting locations.
10662 : */
10663 4 : gotUniqueStartpoint = true;
10664 : }
10665 :
10666 : /*
10667 : * If two base backups are started at the same time (in WAL sender
10668 : * processes), we need to make sure that they use different
10669 : * checkpoints as starting locations, because we use the starting
10670 : * WAL location as a unique identifier for the base backup in the
10671 : * end-of-backup WAL record and when we write the backup history
10672 : * file. Perhaps it would be better generate a separate unique ID
10673 : * for each backup instead of forcing another checkpoint, but
10674 : * taking a checkpoint right after another is not that expensive
10675 : * either because only few buffers have been dirtied yet.
10676 : */
10677 146 : WALInsertLockAcquireExclusive();
10678 146 : if (XLogCtl->Insert.lastBackupStart < startpoint)
10679 : {
10680 146 : XLogCtl->Insert.lastBackupStart = startpoint;
10681 146 : gotUniqueStartpoint = true;
10682 : }
10683 146 : WALInsertLockRelease();
10684 146 : } while (!gotUniqueStartpoint);
10685 :
10686 146 : XLByteToSeg(startpoint, _logSegNo, wal_segment_size);
10687 146 : XLogFileName(xlogfilename, starttli, _logSegNo, wal_segment_size);
10688 :
10689 : /*
10690 : * Construct tablespace_map file
10691 : */
10692 146 : if (exclusive)
10693 2 : tblspcmapfile = makeStringInfo();
10694 :
10695 146 : datadirpathlen = strlen(DataDir);
10696 :
10697 : /*
10698 : * Report that we are now estimating the total backup size if we're
10699 : * streaming base backup as requested by pg_basebackup
10700 : */
10701 146 : if (tablespaces)
10702 144 : pgstat_progress_update_par |
