Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlog.c
4 : * PostgreSQL write-ahead log manager
5 : *
6 : * The Write-Ahead Log (WAL) functionality is split into several source
7 : * files, in addition to this one:
8 : *
9 : * xloginsert.c - Functions for constructing WAL records
10 : * xlogrecovery.c - WAL recovery and standby code
11 : * xlogreader.c - Facility for reading WAL files and parsing WAL records
12 : * xlogutils.c - Helper functions for WAL redo routines
13 : *
14 : * This file contains functions for coordinating database startup and
15 : * checkpointing, and managing the write-ahead log buffers when the
16 : * system is running.
17 : *
18 : * StartupXLOG() is the main entry point of the startup process. It
19 : * coordinates database startup, performing WAL recovery, and the
20 : * transition from WAL recovery into normal operations.
21 : *
22 : * XLogInsertRecord() inserts a WAL record into the WAL buffers. Most
23 : * callers should not call this directly, but use the functions in
24 : * xloginsert.c to construct the WAL record. XLogFlush() can be used
25 : * to force the WAL to disk.
26 : *
27 : * In addition to those, there are many other functions for interrogating
28 : * the current system state, and for starting/stopping backups.
29 : *
30 : *
31 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
32 : * Portions Copyright (c) 1994, Regents of the University of California
33 : *
34 : * src/backend/access/transam/xlog.c
35 : *
36 : *-------------------------------------------------------------------------
37 : */
38 :
39 : #include "postgres.h"
40 :
41 : #include <ctype.h>
42 : #include <math.h>
43 : #include <time.h>
44 : #include <fcntl.h>
45 : #include <sys/stat.h>
46 : #include <sys/time.h>
47 : #include <unistd.h>
48 :
49 : #include "access/clog.h"
50 : #include "access/commit_ts.h"
51 : #include "access/heaptoast.h"
52 : #include "access/multixact.h"
53 : #include "access/rewriteheap.h"
54 : #include "access/subtrans.h"
55 : #include "access/timeline.h"
56 : #include "access/transam.h"
57 : #include "access/twophase.h"
58 : #include "access/xact.h"
59 : #include "access/xlog_internal.h"
60 : #include "access/xlogarchive.h"
61 : #include "access/xloginsert.h"
62 : #include "access/xlogreader.h"
63 : #include "access/xlogrecovery.h"
64 : #include "access/xlogutils.h"
65 : #include "access/xlogwait.h"
66 : #include "backup/basebackup.h"
67 : #include "catalog/catversion.h"
68 : #include "catalog/pg_control.h"
69 : #include "catalog/pg_database.h"
70 : #include "common/controldata_utils.h"
71 : #include "common/file_utils.h"
72 : #include "executor/instrument.h"
73 : #include "miscadmin.h"
74 : #include "pg_trace.h"
75 : #include "pgstat.h"
76 : #include "port/atomics.h"
77 : #include "postmaster/bgwriter.h"
78 : #include "postmaster/startup.h"
79 : #include "postmaster/walsummarizer.h"
80 : #include "postmaster/walwriter.h"
81 : #include "replication/origin.h"
82 : #include "replication/slot.h"
83 : #include "replication/snapbuild.h"
84 : #include "replication/walreceiver.h"
85 : #include "replication/walsender.h"
86 : #include "storage/bufmgr.h"
87 : #include "storage/fd.h"
88 : #include "storage/ipc.h"
89 : #include "storage/large_object.h"
90 : #include "storage/latch.h"
91 : #include "storage/predicate.h"
92 : #include "storage/proc.h"
93 : #include "storage/procarray.h"
94 : #include "storage/reinit.h"
95 : #include "storage/spin.h"
96 : #include "storage/sync.h"
97 : #include "utils/guc_hooks.h"
98 : #include "utils/guc_tables.h"
99 : #include "utils/injection_point.h"
100 : #include "utils/pgstat_internal.h"
101 : #include "utils/ps_status.h"
102 : #include "utils/relmapper.h"
103 : #include "utils/snapmgr.h"
104 : #include "utils/timeout.h"
105 : #include "utils/timestamp.h"
106 : #include "utils/varlena.h"
107 :
108 : #ifdef WAL_DEBUG
109 : #include "utils/memutils.h"
110 : #endif
111 :
112 : /* timeline ID to be used when bootstrapping */
113 : #define BootstrapTimeLineID 1
114 :
115 : /* User-settable parameters */
116 : int max_wal_size_mb = 1024; /* 1 GB */
117 : int min_wal_size_mb = 80; /* 80 MB */
118 : int wal_keep_size_mb = 0;
119 : int XLOGbuffers = -1;
120 : int XLogArchiveTimeout = 0;
121 : int XLogArchiveMode = ARCHIVE_MODE_OFF;
122 : char *XLogArchiveCommand = NULL;
123 : bool EnableHotStandby = false;
124 : bool fullPageWrites = true;
125 : bool wal_log_hints = false;
126 : int wal_compression = WAL_COMPRESSION_NONE;
127 : char *wal_consistency_checking_string = NULL;
128 : bool *wal_consistency_checking = NULL;
129 : bool wal_init_zero = true;
130 : bool wal_recycle = true;
131 : bool log_checkpoints = true;
132 : int wal_sync_method = DEFAULT_WAL_SYNC_METHOD;
133 : int wal_level = WAL_LEVEL_REPLICA;
134 : int CommitDelay = 0; /* precommit delay in microseconds */
135 : int CommitSiblings = 5; /* # concurrent xacts needed to sleep */
136 : int wal_retrieve_retry_interval = 5000;
137 : int max_slot_wal_keep_size_mb = -1;
138 : int wal_decode_buffer_size = 512 * 1024;
139 : bool track_wal_io_timing = false;
140 :
141 : #ifdef WAL_DEBUG
142 : bool XLOG_DEBUG = false;
143 : #endif
144 :
145 : int wal_segment_size = DEFAULT_XLOG_SEG_SIZE;
146 :
147 : /*
148 : * Number of WAL insertion locks to use. A higher value allows more insertions
149 : * to happen concurrently, but adds some CPU overhead to flushing the WAL,
150 : * which needs to iterate all the locks.
151 : */
152 : #define NUM_XLOGINSERT_LOCKS 8
153 :
154 : /*
155 : * Max distance from last checkpoint, before triggering a new xlog-based
156 : * checkpoint.
157 : */
158 : int CheckPointSegments;
159 :
160 : /* Estimated distance between checkpoints, in bytes */
161 : static double CheckPointDistanceEstimate = 0;
162 : static double PrevCheckPointDistance = 0;
163 :
164 : /*
165 : * Track whether there were any deferred checks for custom resource managers
166 : * specified in wal_consistency_checking.
167 : */
168 : static bool check_wal_consistency_checking_deferred = false;
169 :
170 : /*
171 : * GUC support
172 : */
173 : const struct config_enum_entry wal_sync_method_options[] = {
174 : {"fsync", WAL_SYNC_METHOD_FSYNC, false},
175 : #ifdef HAVE_FSYNC_WRITETHROUGH
176 : {"fsync_writethrough", WAL_SYNC_METHOD_FSYNC_WRITETHROUGH, false},
177 : #endif
178 : {"fdatasync", WAL_SYNC_METHOD_FDATASYNC, false},
179 : #ifdef O_SYNC
180 : {"open_sync", WAL_SYNC_METHOD_OPEN, false},
181 : #endif
182 : #ifdef O_DSYNC
183 : {"open_datasync", WAL_SYNC_METHOD_OPEN_DSYNC, false},
184 : #endif
185 : {NULL, 0, false}
186 : };
187 :
188 :
189 : /*
190 : * Although only "on", "off", and "always" are documented,
191 : * we accept all the likely variants of "on" and "off".
192 : */
193 : const struct config_enum_entry archive_mode_options[] = {
194 : {"always", ARCHIVE_MODE_ALWAYS, false},
195 : {"on", ARCHIVE_MODE_ON, false},
196 : {"off", ARCHIVE_MODE_OFF, false},
197 : {"true", ARCHIVE_MODE_ON, true},
198 : {"false", ARCHIVE_MODE_OFF, true},
199 : {"yes", ARCHIVE_MODE_ON, true},
200 : {"no", ARCHIVE_MODE_OFF, true},
201 : {"1", ARCHIVE_MODE_ON, true},
202 : {"0", ARCHIVE_MODE_OFF, true},
203 : {NULL, 0, false}
204 : };
205 :
206 : /*
207 : * Statistics for current checkpoint are collected in this global struct.
208 : * Because only the checkpointer or a stand-alone backend can perform
209 : * checkpoints, this will be unused in normal backends.
210 : */
211 : CheckpointStatsData CheckpointStats;
212 :
213 : /*
214 : * During recovery, lastFullPageWrites keeps track of full_page_writes that
215 : * the replayed WAL records indicate. It's initialized with full_page_writes
216 : * that the recovery starting checkpoint record indicates, and then updated
217 : * each time XLOG_FPW_CHANGE record is replayed.
218 : */
219 : static bool lastFullPageWrites;
220 :
221 : /*
222 : * Local copy of the state tracked by SharedRecoveryState in shared memory,
223 : * It is false if SharedRecoveryState is RECOVERY_STATE_DONE. True actually
224 : * means "not known, need to check the shared state".
225 : */
226 : static bool LocalRecoveryInProgress = true;
227 :
228 : /*
229 : * Local state for XLogInsertAllowed():
230 : * 1: unconditionally allowed to insert XLOG
231 : * 0: unconditionally not allowed to insert XLOG
232 : * -1: must check RecoveryInProgress(); disallow until it is false
233 : * Most processes start with -1 and transition to 1 after seeing that recovery
234 : * is not in progress. But we can also force the value for special cases.
235 : * The coding in XLogInsertAllowed() depends on the first two of these states
236 : * being numerically the same as bool true and false.
237 : */
238 : static int LocalXLogInsertAllowed = -1;
239 :
240 : /*
241 : * ProcLastRecPtr points to the start of the last XLOG record inserted by the
242 : * current backend. It is updated for all inserts. XactLastRecEnd points to
243 : * end+1 of the last record, and is reset when we end a top-level transaction,
244 : * or start a new one; so it can be used to tell if the current transaction has
245 : * created any XLOG records.
246 : *
247 : * While in parallel mode, this may not be fully up to date. When committing,
248 : * a transaction can assume this covers all xlog records written either by the
249 : * user backend or by any parallel worker which was present at any point during
250 : * the transaction. But when aborting, or when still in parallel mode, other
251 : * parallel backends may have written WAL records at later LSNs than the value
252 : * stored here. The parallel leader advances its own copy, when necessary,
253 : * in WaitForParallelWorkersToFinish.
254 : */
255 : XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr;
256 : XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr;
257 : XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr;
258 :
259 : /*
260 : * RedoRecPtr is this backend's local copy of the REDO record pointer
261 : * (which is almost but not quite the same as a pointer to the most recent
262 : * CHECKPOINT record). We update this from the shared-memory copy,
263 : * XLogCtl->Insert.RedoRecPtr, whenever we can safely do so (ie, when we
264 : * hold an insertion lock). See XLogInsertRecord for details. We are also
265 : * allowed to update from XLogCtl->RedoRecPtr if we hold the info_lck;
266 : * see GetRedoRecPtr.
267 : *
268 : * NB: Code that uses this variable must be prepared not only for the
269 : * possibility that it may be arbitrarily out of date, but also for the
270 : * possibility that it might be set to InvalidXLogRecPtr. We used to
271 : * initialize it as a side effect of the first call to RecoveryInProgress(),
272 : * which meant that most code that might use it could assume that it had a
273 : * real if perhaps stale value. That's no longer the case.
274 : */
275 : static XLogRecPtr RedoRecPtr;
276 :
277 : /*
278 : * doPageWrites is this backend's local copy of (fullPageWrites ||
279 : * runningBackups > 0). It is used together with RedoRecPtr to decide whether
280 : * a full-page image of a page need to be taken.
281 : *
282 : * NB: Initially this is false, and there's no guarantee that it will be
283 : * initialized to any other value before it is first used. Any code that
284 : * makes use of it must recheck the value after obtaining a WALInsertLock,
285 : * and respond appropriately if it turns out that the previous value wasn't
286 : * accurate.
287 : */
288 : static bool doPageWrites;
289 :
290 : /*----------
291 : * Shared-memory data structures for XLOG control
292 : *
293 : * LogwrtRqst indicates a byte position that we need to write and/or fsync
294 : * the log up to (all records before that point must be written or fsynced).
295 : * The positions already written/fsynced are maintained in logWriteResult
296 : * and logFlushResult using atomic access.
297 : * In addition to the shared variable, each backend has a private copy of
298 : * both in LogwrtResult, which is updated when convenient.
299 : *
300 : * The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
301 : * (protected by info_lck), but we don't need to cache any copies of it.
302 : *
303 : * info_lck is only held long enough to read/update the protected variables,
304 : * so it's a plain spinlock. The other locks are held longer (potentially
305 : * over I/O operations), so we use LWLocks for them. These locks are:
306 : *
307 : * WALBufMappingLock: must be held to replace a page in the WAL buffer cache.
308 : * It is only held while initializing and changing the mapping. If the
309 : * contents of the buffer being replaced haven't been written yet, the mapping
310 : * lock is released while the write is done, and reacquired afterwards.
311 : *
312 : * WALWriteLock: must be held to write WAL buffers to disk (XLogWrite or
313 : * XLogFlush).
314 : *
315 : * ControlFileLock: must be held to read/update control file or create
316 : * new log file.
317 : *
318 : *----------
319 : */
320 :
321 : typedef struct XLogwrtRqst
322 : {
323 : XLogRecPtr Write; /* last byte + 1 to write out */
324 : XLogRecPtr Flush; /* last byte + 1 to flush */
325 : } XLogwrtRqst;
326 :
327 : typedef struct XLogwrtResult
328 : {
329 : XLogRecPtr Write; /* last byte + 1 written out */
330 : XLogRecPtr Flush; /* last byte + 1 flushed */
331 : } XLogwrtResult;
332 :
333 : /*
334 : * Inserting to WAL is protected by a small fixed number of WAL insertion
335 : * locks. To insert to the WAL, you must hold one of the locks - it doesn't
336 : * matter which one. To lock out other concurrent insertions, you must hold
337 : * of them. Each WAL insertion lock consists of a lightweight lock, plus an
338 : * indicator of how far the insertion has progressed (insertingAt).
339 : *
340 : * The insertingAt values are read when a process wants to flush WAL from
341 : * the in-memory buffers to disk, to check that all the insertions to the
342 : * region the process is about to write out have finished. You could simply
343 : * wait for all currently in-progress insertions to finish, but the
344 : * insertingAt indicator allows you to ignore insertions to later in the WAL,
345 : * so that you only wait for the insertions that are modifying the buffers
346 : * you're about to write out.
347 : *
348 : * This isn't just an optimization. If all the WAL buffers are dirty, an
349 : * inserter that's holding a WAL insert lock might need to evict an old WAL
350 : * buffer, which requires flushing the WAL. If it's possible for an inserter
351 : * to block on another inserter unnecessarily, deadlock can arise when two
352 : * inserters holding a WAL insert lock wait for each other to finish their
353 : * insertion.
354 : *
355 : * Small WAL records that don't cross a page boundary never update the value,
356 : * the WAL record is just copied to the page and the lock is released. But
357 : * to avoid the deadlock-scenario explained above, the indicator is always
358 : * updated before sleeping while holding an insertion lock.
359 : *
360 : * lastImportantAt contains the LSN of the last important WAL record inserted
361 : * using a given lock. This value is used to detect if there has been
362 : * important WAL activity since the last time some action, like a checkpoint,
363 : * was performed - allowing to not repeat the action if not. The LSN is
364 : * updated for all insertions, unless the XLOG_MARK_UNIMPORTANT flag was
365 : * set. lastImportantAt is never cleared, only overwritten by the LSN of newer
366 : * records. Tracking the WAL activity directly in WALInsertLock has the
367 : * advantage of not needing any additional locks to update the value.
368 : */
369 : typedef struct
370 : {
371 : LWLock lock;
372 : pg_atomic_uint64 insertingAt;
373 : XLogRecPtr lastImportantAt;
374 : } WALInsertLock;
375 :
376 : /*
377 : * All the WAL insertion locks are allocated as an array in shared memory. We
378 : * force the array stride to be a power of 2, which saves a few cycles in
379 : * indexing, but more importantly also ensures that individual slots don't
380 : * cross cache line boundaries. (Of course, we have to also ensure that the
381 : * array start address is suitably aligned.)
382 : */
383 : typedef union WALInsertLockPadded
384 : {
385 : WALInsertLock l;
386 : char pad[PG_CACHE_LINE_SIZE];
387 : } WALInsertLockPadded;
388 :
389 : /*
390 : * Session status of running backup, used for sanity checks in SQL-callable
391 : * functions to start and stop backups.
392 : */
393 : static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE;
394 :
395 : /*
396 : * Shared state data for WAL insertion.
397 : */
398 : typedef struct XLogCtlInsert
399 : {
400 : slock_t insertpos_lck; /* protects CurrBytePos and PrevBytePos */
401 :
402 : /*
403 : * CurrBytePos is the end of reserved WAL. The next record will be
404 : * inserted at that position. PrevBytePos is the start position of the
405 : * previously inserted (or rather, reserved) record - it is copied to the
406 : * prev-link of the next record. These are stored as "usable byte
407 : * positions" rather than XLogRecPtrs (see XLogBytePosToRecPtr()).
408 : */
409 : uint64 CurrBytePos;
410 : uint64 PrevBytePos;
411 :
412 : /*
413 : * Make sure the above heavily-contended spinlock and byte positions are
414 : * on their own cache line. In particular, the RedoRecPtr and full page
415 : * write variables below should be on a different cache line. They are
416 : * read on every WAL insertion, but updated rarely, and we don't want
417 : * those reads to steal the cache line containing Curr/PrevBytePos.
418 : */
419 : char pad[PG_CACHE_LINE_SIZE];
420 :
421 : /*
422 : * fullPageWrites is the authoritative value used by all backends to
423 : * determine whether to write full-page image to WAL. This shared value,
424 : * instead of the process-local fullPageWrites, is required because, when
425 : * full_page_writes is changed by SIGHUP, we must WAL-log it before it
426 : * actually affects WAL-logging by backends. Checkpointer sets at startup
427 : * or after SIGHUP.
428 : *
429 : * To read these fields, you must hold an insertion lock. To modify them,
430 : * you must hold ALL the locks.
431 : */
432 : XLogRecPtr RedoRecPtr; /* current redo point for insertions */
433 : bool fullPageWrites;
434 :
435 : /*
436 : * runningBackups is a counter indicating the number of backups currently
437 : * in progress. lastBackupStart is the latest checkpoint redo location
438 : * used as a starting point for an online backup.
439 : */
440 : int runningBackups;
441 : XLogRecPtr lastBackupStart;
442 :
443 : /*
444 : * WAL insertion locks.
445 : */
446 : WALInsertLockPadded *WALInsertLocks;
447 : } XLogCtlInsert;
448 :
449 : /*
450 : * Total shared-memory state for XLOG.
451 : */
452 : typedef struct XLogCtlData
453 : {
454 : XLogCtlInsert Insert;
455 :
456 : /* Protected by info_lck: */
457 : XLogwrtRqst LogwrtRqst;
458 : XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
459 : XLogRecPtr asyncXactLSN; /* LSN of newest async commit/abort */
460 : XLogRecPtr replicationSlotMinLSN; /* oldest LSN needed by any slot */
461 :
462 : XLogSegNo lastRemovedSegNo; /* latest removed/recycled XLOG segment */
463 :
464 : /* Fake LSN counter, for unlogged relations. */
465 : pg_atomic_uint64 unloggedLSN;
466 :
467 : /* Time and LSN of last xlog segment switch. Protected by WALWriteLock. */
468 : pg_time_t lastSegSwitchTime;
469 : XLogRecPtr lastSegSwitchLSN;
470 :
471 : /* These are accessed using atomics -- info_lck not needed */
472 : pg_atomic_uint64 logInsertResult; /* last byte + 1 inserted to buffers */
473 : pg_atomic_uint64 logWriteResult; /* last byte + 1 written out */
474 : pg_atomic_uint64 logFlushResult; /* last byte + 1 flushed */
475 :
476 : /*
477 : * Latest initialized page in the cache (last byte position + 1).
478 : *
479 : * To change the identity of a buffer (and InitializedUpTo), you need to
480 : * hold WALBufMappingLock. To change the identity of a buffer that's
481 : * still dirty, the old page needs to be written out first, and for that
482 : * you need WALWriteLock, and you need to ensure that there are no
483 : * in-progress insertions to the page by calling
484 : * WaitXLogInsertionsToFinish().
485 : */
486 : XLogRecPtr InitializedUpTo;
487 :
488 : /*
489 : * These values do not change after startup, although the pointed-to pages
490 : * and xlblocks values certainly do. xlblocks values are protected by
491 : * WALBufMappingLock.
492 : */
493 : char *pages; /* buffers for unwritten XLOG pages */
494 : pg_atomic_uint64 *xlblocks; /* 1st byte ptr-s + XLOG_BLCKSZ */
495 : int XLogCacheBlck; /* highest allocated xlog buffer index */
496 :
497 : /*
498 : * InsertTimeLineID is the timeline into which new WAL is being inserted
499 : * and flushed. It is zero during recovery, and does not change once set.
500 : *
501 : * If we create a new timeline when the system was started up,
502 : * PrevTimeLineID is the old timeline's ID that we forked off from.
503 : * Otherwise it's equal to InsertTimeLineID.
504 : *
505 : * We set these fields while holding info_lck. Most that reads these
506 : * values knows that recovery is no longer in progress and so can safely
507 : * read the value without a lock, but code that could be run either during
508 : * or after recovery can take info_lck while reading these values.
509 : */
510 : TimeLineID InsertTimeLineID;
511 : TimeLineID PrevTimeLineID;
512 :
513 : /*
514 : * SharedRecoveryState indicates if we're still in crash or archive
515 : * recovery. Protected by info_lck.
516 : */
517 : RecoveryState SharedRecoveryState;
518 :
519 : /*
520 : * InstallXLogFileSegmentActive indicates whether the checkpointer should
521 : * arrange for future segments by recycling and/or PreallocXlogFiles().
522 : * Protected by ControlFileLock. Only the startup process changes it. If
523 : * true, anyone can use InstallXLogFileSegment(). If false, the startup
524 : * process owns the exclusive right to install segments, by reading from
525 : * the archive and possibly replacing existing files.
526 : */
527 : bool InstallXLogFileSegmentActive;
528 :
529 : /*
530 : * WalWriterSleeping indicates whether the WAL writer is currently in
531 : * low-power mode (and hence should be nudged if an async commit occurs).
532 : * Protected by info_lck.
533 : */
534 : bool WalWriterSleeping;
535 :
536 : /*
537 : * During recovery, we keep a copy of the latest checkpoint record here.
538 : * lastCheckPointRecPtr points to start of checkpoint record and
539 : * lastCheckPointEndPtr points to end+1 of checkpoint record. Used by the
540 : * checkpointer when it wants to create a restartpoint.
541 : *
542 : * Protected by info_lck.
543 : */
544 : XLogRecPtr lastCheckPointRecPtr;
545 : XLogRecPtr lastCheckPointEndPtr;
546 : CheckPoint lastCheckPoint;
547 :
548 : /*
549 : * lastFpwDisableRecPtr points to the start of the last replayed
550 : * XLOG_FPW_CHANGE record that instructs full_page_writes is disabled.
551 : */
552 : XLogRecPtr lastFpwDisableRecPtr;
553 :
554 : slock_t info_lck; /* locks shared variables shown above */
555 : } XLogCtlData;
556 :
557 : /*
558 : * Classification of XLogInsertRecord operations.
559 : */
560 : typedef enum
561 : {
562 : WALINSERT_NORMAL,
563 : WALINSERT_SPECIAL_SWITCH,
564 : WALINSERT_SPECIAL_CHECKPOINT
565 : } WalInsertClass;
566 :
567 : static XLogCtlData *XLogCtl = NULL;
568 :
569 : /* a private copy of XLogCtl->Insert.WALInsertLocks, for convenience */
570 : static WALInsertLockPadded *WALInsertLocks = NULL;
571 :
572 : /*
573 : * We maintain an image of pg_control in shared memory.
574 : */
575 : static ControlFileData *ControlFile = NULL;
576 :
577 : /*
578 : * Calculate the amount of space left on the page after 'endptr'. Beware
579 : * multiple evaluation!
580 : */
581 : #define INSERT_FREESPACE(endptr) \
582 : (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
583 :
584 : /* Macro to advance to next buffer index. */
585 : #define NextBufIdx(idx) \
586 : (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
587 :
588 : /*
589 : * XLogRecPtrToBufIdx returns the index of the WAL buffer that holds, or
590 : * would hold if it was in cache, the page containing 'recptr'.
591 : */
592 : #define XLogRecPtrToBufIdx(recptr) \
593 : (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
594 :
595 : /*
596 : * These are the number of bytes in a WAL page usable for WAL data.
597 : */
598 : #define UsableBytesInPage (XLOG_BLCKSZ - SizeOfXLogShortPHD)
599 :
600 : /*
601 : * Convert values of GUCs measured in megabytes to equiv. segment count.
602 : * Rounds down.
603 : */
604 : #define ConvertToXSegs(x, segsize) XLogMBVarToSegs((x), (segsize))
605 :
606 : /* The number of bytes in a WAL segment usable for WAL data. */
607 : static int UsableBytesInSegment;
608 :
609 : /*
610 : * Private, possibly out-of-date copy of shared LogwrtResult.
611 : * See discussion above.
612 : */
613 : static XLogwrtResult LogwrtResult = {0, 0};
614 :
615 : /*
616 : * Update local copy of shared XLogCtl->log{Write,Flush}Result
617 : *
618 : * It's critical that Flush always trails Write, so the order of the reads is
619 : * important, as is the barrier. See also XLogWrite.
620 : */
621 : #define RefreshXLogWriteResult(_target) \
622 : do { \
623 : _target.Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult); \
624 : pg_read_barrier(); \
625 : _target.Write = pg_atomic_read_u64(&XLogCtl->logWriteResult); \
626 : } while (0)
627 :
628 : /*
629 : * openLogFile is -1 or a kernel FD for an open log file segment.
630 : * openLogSegNo identifies the segment, and openLogTLI the corresponding TLI.
631 : * These variables are only used to write the XLOG, and so will normally refer
632 : * to the active segment.
633 : *
634 : * Note: call Reserve/ReleaseExternalFD to track consumption of this FD.
635 : */
636 : static int openLogFile = -1;
637 : static XLogSegNo openLogSegNo = 0;
638 : static TimeLineID openLogTLI = 0;
639 :
640 : /*
641 : * Local copies of equivalent fields in the control file. When running
642 : * crash recovery, LocalMinRecoveryPoint is set to InvalidXLogRecPtr as we
643 : * expect to replay all the WAL available, and updateMinRecoveryPoint is
644 : * switched to false to prevent any updates while replaying records.
645 : * Those values are kept consistent as long as crash recovery runs.
646 : */
647 : static XLogRecPtr LocalMinRecoveryPoint;
648 : static TimeLineID LocalMinRecoveryPointTLI;
649 : static bool updateMinRecoveryPoint = true;
650 :
651 : /* For WALInsertLockAcquire/Release functions */
652 : static int MyLockNo = 0;
653 : static bool holdingAllLocks = false;
654 :
655 : #ifdef WAL_DEBUG
656 : static MemoryContext walDebugCxt = NULL;
657 : #endif
658 :
659 : static void CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI,
660 : XLogRecPtr EndOfLog,
661 : TimeLineID newTLI);
662 : static void CheckRequiredParameterValues(void);
663 : static void XLogReportParameters(void);
664 : static int LocalSetXLogInsertAllowed(void);
665 : static void CreateEndOfRecoveryRecord(void);
666 : static XLogRecPtr CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn,
667 : XLogRecPtr pagePtr,
668 : TimeLineID newTLI);
669 : static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags);
670 : static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo);
671 : static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void);
672 :
673 : static void AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli,
674 : bool opportunistic);
675 : static void XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible);
676 : static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
677 : bool find_free, XLogSegNo max_segno,
678 : TimeLineID tli);
679 : static void XLogFileClose(void);
680 : static void PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli);
681 : static void RemoveTempXlogFiles(void);
682 : static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr,
683 : XLogRecPtr endptr, TimeLineID insertTLI);
684 : static void RemoveXlogFile(const struct dirent *segment_de,
685 : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
686 : TimeLineID insertTLI);
687 : static void UpdateLastRemovedPtr(char *filename);
688 : static void ValidateXLOGDirectoryStructure(void);
689 : static void CleanupBackupHistory(void);
690 : static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force);
691 : static bool PerformRecoveryXLogAction(void);
692 : static void InitControlFile(uint64 sysidentifier, uint32 data_checksum_version);
693 : static void WriteControlFile(void);
694 : static void ReadControlFile(void);
695 : static void UpdateControlFile(void);
696 : static char *str_time(pg_time_t tnow, char *buf, size_t bufsize);
697 :
698 : static int get_sync_bit(int method);
699 :
700 : static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch,
701 : XLogRecData *rdata,
702 : XLogRecPtr StartPos, XLogRecPtr EndPos,
703 : TimeLineID tli);
704 : static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos,
705 : XLogRecPtr *EndPos, XLogRecPtr *PrevPtr);
706 : static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos,
707 : XLogRecPtr *PrevPtr);
708 : static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto);
709 : static char *GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli);
710 : static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos);
711 : static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos);
712 : static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr);
713 :
714 : static void WALInsertLockAcquire(void);
715 : static void WALInsertLockAcquireExclusive(void);
716 : static void WALInsertLockRelease(void);
717 : static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
718 :
719 : /*
720 : * Insert an XLOG record represented by an already-constructed chain of data
721 : * chunks. This is a low-level routine; to construct the WAL record header
722 : * and data, use the higher-level routines in xloginsert.c.
723 : *
724 : * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
725 : * WAL record applies to, that were not included in the record as full page
726 : * images. If fpw_lsn <= RedoRecPtr, the function does not perform the
727 : * insertion and returns InvalidXLogRecPtr. The caller can then recalculate
728 : * which pages need a full-page image, and retry. If fpw_lsn is invalid, the
729 : * record is always inserted.
730 : *
731 : * 'flags' gives more in-depth control on the record being inserted. See
732 : * XLogSetRecordFlags() for details.
733 : *
734 : * 'topxid_included' tells whether the top-transaction id is logged along with
735 : * current subtransaction. See XLogRecordAssemble().
736 : *
737 : * The first XLogRecData in the chain must be for the record header, and its
738 : * data must be MAXALIGNed. XLogInsertRecord fills in the xl_prev and
739 : * xl_crc fields in the header, the rest of the header must already be filled
740 : * by the caller.
741 : *
742 : * Returns XLOG pointer to end of record (beginning of next record).
743 : * This can be used as LSN for data pages affected by the logged action.
744 : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
745 : * before the data page can be written out. This implements the basic
746 : * WAL rule "write the log before the data".)
747 : */
748 : XLogRecPtr
749 29247234 : XLogInsertRecord(XLogRecData *rdata,
750 : XLogRecPtr fpw_lsn,
751 : uint8 flags,
752 : int num_fpi,
753 : uint64 fpi_bytes,
754 : bool topxid_included)
755 : {
756 29247234 : XLogCtlInsert *Insert = &XLogCtl->Insert;
757 : pg_crc32c rdata_crc;
758 : bool inserted;
759 29247234 : XLogRecord *rechdr = (XLogRecord *) rdata->data;
760 29247234 : uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
761 29247234 : WalInsertClass class = WALINSERT_NORMAL;
762 : XLogRecPtr StartPos;
763 : XLogRecPtr EndPos;
764 29247234 : bool prevDoPageWrites = doPageWrites;
765 : TimeLineID insertTLI;
766 :
767 : /* Does this record type require special handling? */
768 29247234 : if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
769 : {
770 447756 : if (info == XLOG_SWITCH)
771 1566 : class = WALINSERT_SPECIAL_SWITCH;
772 446190 : else if (info == XLOG_CHECKPOINT_REDO)
773 1816 : class = WALINSERT_SPECIAL_CHECKPOINT;
774 : }
775 :
776 : /* we assume that all of the record header is in the first chunk */
777 : Assert(rdata->len >= SizeOfXLogRecord);
778 :
779 : /* cross-check on whether we should be here or not */
780 29247234 : if (!XLogInsertAllowed())
781 0 : elog(ERROR, "cannot make new WAL entries during recovery");
782 :
783 : /*
784 : * Given that we're not in recovery, InsertTimeLineID is set and can't
785 : * change, so we can read it without a lock.
786 : */
787 29247234 : insertTLI = XLogCtl->InsertTimeLineID;
788 :
789 : /*----------
790 : *
791 : * We have now done all the preparatory work we can without holding a
792 : * lock or modifying shared state. From here on, inserting the new WAL
793 : * record to the shared WAL buffer cache is a two-step process:
794 : *
795 : * 1. Reserve the right amount of space from the WAL. The current head of
796 : * reserved space is kept in Insert->CurrBytePos, and is protected by
797 : * insertpos_lck.
798 : *
799 : * 2. Copy the record to the reserved WAL space. This involves finding the
800 : * correct WAL buffer containing the reserved space, and copying the
801 : * record in place. This can be done concurrently in multiple processes.
802 : *
803 : * To keep track of which insertions are still in-progress, each concurrent
804 : * inserter acquires an insertion lock. In addition to just indicating that
805 : * an insertion is in progress, the lock tells others how far the inserter
806 : * has progressed. There is a small fixed number of insertion locks,
807 : * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
808 : * boundary, it updates the value stored in the lock to the how far it has
809 : * inserted, to allow the previous buffer to be flushed.
810 : *
811 : * Holding onto an insertion lock also protects RedoRecPtr and
812 : * fullPageWrites from changing until the insertion is finished.
813 : *
814 : * Step 2 can usually be done completely in parallel. If the required WAL
815 : * page is not initialized yet, you have to grab WALBufMappingLock to
816 : * initialize it, but the WAL writer tries to do that ahead of insertions
817 : * to avoid that from happening in the critical path.
818 : *
819 : *----------
820 : */
821 29247234 : START_CRIT_SECTION();
822 :
823 29247234 : if (likely(class == WALINSERT_NORMAL))
824 : {
825 29243852 : WALInsertLockAcquire();
826 :
827 : /*
828 : * Check to see if my copy of RedoRecPtr is out of date. If so, may
829 : * have to go back and have the caller recompute everything. This can
830 : * only happen just after a checkpoint, so it's better to be slow in
831 : * this case and fast otherwise.
832 : *
833 : * Also check to see if fullPageWrites was just turned on or there's a
834 : * running backup (which forces full-page writes); if we weren't
835 : * already doing full-page writes then go back and recompute.
836 : *
837 : * If we aren't doing full-page writes then RedoRecPtr doesn't
838 : * actually affect the contents of the XLOG record, so we'll update
839 : * our local copy but not force a recomputation. (If doPageWrites was
840 : * just turned off, we could recompute the record without full pages,
841 : * but we choose not to bother.)
842 : */
843 29243852 : if (RedoRecPtr != Insert->RedoRecPtr)
844 : {
845 : Assert(RedoRecPtr < Insert->RedoRecPtr);
846 13788 : RedoRecPtr = Insert->RedoRecPtr;
847 : }
848 29243852 : doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
849 :
850 29243852 : if (doPageWrites &&
851 28758858 : (!prevDoPageWrites ||
852 26392266 : (XLogRecPtrIsValid(fpw_lsn) && fpw_lsn <= RedoRecPtr)))
853 : {
854 : /*
855 : * Oops, some buffer now needs to be backed up that the caller
856 : * didn't back up. Start over.
857 : */
858 15358 : WALInsertLockRelease();
859 15358 : END_CRIT_SECTION();
860 15358 : return InvalidXLogRecPtr;
861 : }
862 :
863 : /*
864 : * Reserve space for the record in the WAL. This also sets the xl_prev
865 : * pointer.
866 : */
867 29228494 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
868 : &rechdr->xl_prev);
869 :
870 : /* Normal records are always inserted. */
871 29228494 : inserted = true;
872 : }
873 3382 : else if (class == WALINSERT_SPECIAL_SWITCH)
874 : {
875 : /*
876 : * In order to insert an XLOG_SWITCH record, we need to hold all of
877 : * the WAL insertion locks, not just one, so that no one else can
878 : * begin inserting a record until we've figured out how much space
879 : * remains in the current WAL segment and claimed all of it.
880 : *
881 : * Nonetheless, this case is simpler than the normal cases handled
882 : * below, which must check for changes in doPageWrites and RedoRecPtr.
883 : * Those checks are only needed for records that can contain buffer
884 : * references, and an XLOG_SWITCH record never does.
885 : */
886 : Assert(!XLogRecPtrIsValid(fpw_lsn));
887 1566 : WALInsertLockAcquireExclusive();
888 1566 : inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
889 : }
890 : else
891 : {
892 : Assert(class == WALINSERT_SPECIAL_CHECKPOINT);
893 :
894 : /*
895 : * We need to update both the local and shared copies of RedoRecPtr,
896 : * which means that we need to hold all the WAL insertion locks.
897 : * However, there can't be any buffer references, so as above, we need
898 : * not check RedoRecPtr before inserting the record; we just need to
899 : * update it afterwards.
900 : */
901 : Assert(!XLogRecPtrIsValid(fpw_lsn));
902 1816 : WALInsertLockAcquireExclusive();
903 1816 : ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
904 : &rechdr->xl_prev);
905 1816 : RedoRecPtr = Insert->RedoRecPtr = StartPos;
906 1816 : inserted = true;
907 : }
908 :
909 29231876 : if (inserted)
910 : {
911 : /*
912 : * Now that xl_prev has been filled in, calculate CRC of the record
913 : * header.
914 : */
915 29231760 : rdata_crc = rechdr->xl_crc;
916 29231760 : COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
917 29231760 : FIN_CRC32C(rdata_crc);
918 29231760 : rechdr->xl_crc = rdata_crc;
919 :
920 : /*
921 : * All the record data, including the header, is now ready to be
922 : * inserted. Copy the record in the space reserved.
923 : */
924 29231760 : CopyXLogRecordToWAL(rechdr->xl_tot_len,
925 : class == WALINSERT_SPECIAL_SWITCH, rdata,
926 : StartPos, EndPos, insertTLI);
927 :
928 : /*
929 : * Unless record is flagged as not important, update LSN of last
930 : * important record in the current slot. When holding all locks, just
931 : * update the first one.
932 : */
933 29231760 : if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
934 : {
935 29049292 : int lockno = holdingAllLocks ? 0 : MyLockNo;
936 :
937 29049292 : WALInsertLocks[lockno].l.lastImportantAt = StartPos;
938 : }
939 : }
940 : else
941 : {
942 : /*
943 : * This was an xlog-switch record, but the current insert location was
944 : * already exactly at the beginning of a segment, so there was no need
945 : * to do anything.
946 : */
947 : }
948 :
949 : /*
950 : * Done! Let others know that we're finished.
951 : */
952 29231876 : WALInsertLockRelease();
953 :
954 29231876 : END_CRIT_SECTION();
955 :
956 29231876 : MarkCurrentTransactionIdLoggedIfAny();
957 :
958 : /*
959 : * Mark top transaction id is logged (if needed) so that we should not try
960 : * to log it again with the next WAL record in the current subtransaction.
961 : */
962 29231876 : if (topxid_included)
963 438 : MarkSubxactTopXidLogged();
964 :
965 : /*
966 : * Update shared LogwrtRqst.Write, if we crossed page boundary.
967 : */
968 29231876 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
969 : {
970 3326264 : SpinLockAcquire(&XLogCtl->info_lck);
971 : /* advance global request to include new block(s) */
972 3326264 : if (XLogCtl->LogwrtRqst.Write < EndPos)
973 3138020 : XLogCtl->LogwrtRqst.Write = EndPos;
974 3326264 : SpinLockRelease(&XLogCtl->info_lck);
975 3326264 : RefreshXLogWriteResult(LogwrtResult);
976 : }
977 :
978 : /*
979 : * If this was an XLOG_SWITCH record, flush the record and the empty
980 : * padding space that fills the rest of the segment, and perform
981 : * end-of-segment actions (eg, notifying archiver).
982 : */
983 29231876 : if (class == WALINSERT_SPECIAL_SWITCH)
984 : {
985 : TRACE_POSTGRESQL_WAL_SWITCH();
986 1566 : XLogFlush(EndPos);
987 :
988 : /*
989 : * Even though we reserved the rest of the segment for us, which is
990 : * reflected in EndPos, we return a pointer to just the end of the
991 : * xlog-switch record.
992 : */
993 1566 : if (inserted)
994 : {
995 1450 : EndPos = StartPos + SizeOfXLogRecord;
996 1450 : if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
997 : {
998 0 : uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
999 :
1000 0 : if (offset == EndPos % XLOG_BLCKSZ)
1001 0 : EndPos += SizeOfXLogLongPHD;
1002 : else
1003 0 : EndPos += SizeOfXLogShortPHD;
1004 : }
1005 : }
1006 : }
1007 :
1008 : #ifdef WAL_DEBUG
1009 : if (XLOG_DEBUG)
1010 : {
1011 : static XLogReaderState *debug_reader = NULL;
1012 : XLogRecord *record;
1013 : DecodedXLogRecord *decoded;
1014 : StringInfoData buf;
1015 : StringInfoData recordBuf;
1016 : char *errormsg = NULL;
1017 : MemoryContext oldCxt;
1018 :
1019 : oldCxt = MemoryContextSwitchTo(walDebugCxt);
1020 :
1021 : initStringInfo(&buf);
1022 : appendStringInfo(&buf, "INSERT @ %X/%08X: ", LSN_FORMAT_ARGS(EndPos));
1023 :
1024 : /*
1025 : * We have to piece together the WAL record data from the XLogRecData
1026 : * entries, so that we can pass it to the rm_desc function as one
1027 : * contiguous chunk.
1028 : */
1029 : initStringInfo(&recordBuf);
1030 : for (; rdata != NULL; rdata = rdata->next)
1031 : appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1032 :
1033 : /* We also need temporary space to decode the record. */
1034 : record = (XLogRecord *) recordBuf.data;
1035 : decoded = (DecodedXLogRecord *)
1036 : palloc(DecodeXLogRecordRequiredSpace(record->xl_tot_len));
1037 :
1038 : if (!debug_reader)
1039 : debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1040 : XL_ROUTINE(.page_read = NULL,
1041 : .segment_open = NULL,
1042 : .segment_close = NULL),
1043 : NULL);
1044 : if (!debug_reader)
1045 : {
1046 : appendStringInfoString(&buf, "error decoding record: out of memory while allocating a WAL reading processor");
1047 : }
1048 : else if (!DecodeXLogRecord(debug_reader,
1049 : decoded,
1050 : record,
1051 : EndPos,
1052 : &errormsg))
1053 : {
1054 : appendStringInfo(&buf, "error decoding record: %s",
1055 : errormsg ? errormsg : "no error message");
1056 : }
1057 : else
1058 : {
1059 : appendStringInfoString(&buf, " - ");
1060 :
1061 : debug_reader->record = decoded;
1062 : xlog_outdesc(&buf, debug_reader);
1063 : debug_reader->record = NULL;
1064 : }
1065 : elog(LOG, "%s", buf.data);
1066 :
1067 : pfree(decoded);
1068 : pfree(buf.data);
1069 : pfree(recordBuf.data);
1070 : MemoryContextSwitchTo(oldCxt);
1071 : }
1072 : #endif
1073 :
1074 : /*
1075 : * Update our global variables
1076 : */
1077 29231876 : ProcLastRecPtr = StartPos;
1078 29231876 : XactLastRecEnd = EndPos;
1079 :
1080 : /* Report WAL traffic to the instrumentation. */
1081 29231876 : if (inserted)
1082 : {
1083 29231760 : pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1084 29231760 : pgWalUsage.wal_records++;
1085 29231760 : pgWalUsage.wal_fpi += num_fpi;
1086 29231760 : pgWalUsage.wal_fpi_bytes += fpi_bytes;
1087 :
1088 : /* Required for the flush of pending stats WAL data */
1089 29231760 : pgstat_report_fixed = true;
1090 : }
1091 :
1092 29231876 : return EndPos;
1093 : }
1094 :
1095 : /*
1096 : * Reserves the right amount of space for a record of given size from the WAL.
1097 : * *StartPos is set to the beginning of the reserved section, *EndPos to
1098 : * its end+1. *PrevPtr is set to the beginning of the previous record; it is
1099 : * used to set the xl_prev of this record.
1100 : *
1101 : * This is the performance critical part of XLogInsert that must be serialized
1102 : * across backends. The rest can happen mostly in parallel. Try to keep this
1103 : * section as short as possible, insertpos_lck can be heavily contended on a
1104 : * busy system.
1105 : *
1106 : * NB: The space calculation here must match the code in CopyXLogRecordToWAL,
1107 : * where we actually copy the record to the reserved space.
1108 : *
1109 : * NB: Testing shows that XLogInsertRecord runs faster if this code is inlined;
1110 : * however, because there are two call sites, the compiler is reluctant to
1111 : * inline. We use pg_attribute_always_inline here to try to convince it.
1112 : */
1113 : static pg_attribute_always_inline void
1114 29230310 : ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos,
1115 : XLogRecPtr *PrevPtr)
1116 : {
1117 29230310 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1118 : uint64 startbytepos;
1119 : uint64 endbytepos;
1120 : uint64 prevbytepos;
1121 :
1122 29230310 : size = MAXALIGN(size);
1123 :
1124 : /* All (non xlog-switch) records should contain data. */
1125 : Assert(size > SizeOfXLogRecord);
1126 :
1127 : /*
1128 : * The duration the spinlock needs to be held is minimized by minimizing
1129 : * the calculations that have to be done while holding the lock. The
1130 : * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1131 : * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1132 : * page headers. The mapping between "usable" byte positions and physical
1133 : * positions (XLogRecPtrs) can be done outside the locked region, and
1134 : * because the usable byte position doesn't include any headers, reserving
1135 : * X bytes from WAL is almost as simple as "CurrBytePos += X".
1136 : */
1137 29230310 : SpinLockAcquire(&Insert->insertpos_lck);
1138 :
1139 29230310 : startbytepos = Insert->CurrBytePos;
1140 29230310 : endbytepos = startbytepos + size;
1141 29230310 : prevbytepos = Insert->PrevBytePos;
1142 29230310 : Insert->CurrBytePos = endbytepos;
1143 29230310 : Insert->PrevBytePos = startbytepos;
1144 :
1145 29230310 : SpinLockRelease(&Insert->insertpos_lck);
1146 :
1147 29230310 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1148 29230310 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1149 29230310 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1150 :
1151 : /*
1152 : * Check that the conversions between "usable byte positions" and
1153 : * XLogRecPtrs work consistently in both directions.
1154 : */
1155 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1156 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1157 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1158 29230310 : }
1159 :
1160 : /*
1161 : * Like ReserveXLogInsertLocation(), but for an xlog-switch record.
1162 : *
1163 : * A log-switch record is handled slightly differently. The rest of the
1164 : * segment will be reserved for this insertion, as indicated by the returned
1165 : * *EndPos value. However, if we are already at the beginning of the current
1166 : * segment, *StartPos and *EndPos are set to the current location without
1167 : * reserving any space, and the function returns false.
1168 : */
1169 : static bool
1170 1566 : ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
1171 : {
1172 1566 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1173 : uint64 startbytepos;
1174 : uint64 endbytepos;
1175 : uint64 prevbytepos;
1176 1566 : uint32 size = MAXALIGN(SizeOfXLogRecord);
1177 : XLogRecPtr ptr;
1178 : uint32 segleft;
1179 :
1180 : /*
1181 : * These calculations are a bit heavy-weight to be done while holding a
1182 : * spinlock, but since we're holding all the WAL insertion locks, there
1183 : * are no other inserters competing for it. GetXLogInsertRecPtr() does
1184 : * compete for it, but that's not called very frequently.
1185 : */
1186 1566 : SpinLockAcquire(&Insert->insertpos_lck);
1187 :
1188 1566 : startbytepos = Insert->CurrBytePos;
1189 :
1190 1566 : ptr = XLogBytePosToEndRecPtr(startbytepos);
1191 1566 : if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1192 : {
1193 116 : SpinLockRelease(&Insert->insertpos_lck);
1194 116 : *EndPos = *StartPos = ptr;
1195 116 : return false;
1196 : }
1197 :
1198 1450 : endbytepos = startbytepos + size;
1199 1450 : prevbytepos = Insert->PrevBytePos;
1200 :
1201 1450 : *StartPos = XLogBytePosToRecPtr(startbytepos);
1202 1450 : *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1203 :
1204 1450 : segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1205 1450 : if (segleft != wal_segment_size)
1206 : {
1207 : /* consume the rest of the segment */
1208 1450 : *EndPos += segleft;
1209 1450 : endbytepos = XLogRecPtrToBytePos(*EndPos);
1210 : }
1211 1450 : Insert->CurrBytePos = endbytepos;
1212 1450 : Insert->PrevBytePos = startbytepos;
1213 :
1214 1450 : SpinLockRelease(&Insert->insertpos_lck);
1215 :
1216 1450 : *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1217 :
1218 : Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
1219 : Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1220 : Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1221 : Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1222 :
1223 1450 : return true;
1224 : }
1225 :
1226 : /*
1227 : * Subroutine of XLogInsertRecord. Copies a WAL record to an already-reserved
1228 : * area in the WAL.
1229 : */
1230 : static void
1231 29231760 : CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata,
1232 : XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
1233 : {
1234 : char *currpos;
1235 : int freespace;
1236 : int written;
1237 : XLogRecPtr CurrPos;
1238 : XLogPageHeader pagehdr;
1239 :
1240 : /*
1241 : * Get a pointer to the right place in the right WAL buffer to start
1242 : * inserting to.
1243 : */
1244 29231760 : CurrPos = StartPos;
1245 29231760 : currpos = GetXLogBuffer(CurrPos, tli);
1246 29231760 : freespace = INSERT_FREESPACE(CurrPos);
1247 :
1248 : /*
1249 : * there should be enough space for at least the first field (xl_tot_len)
1250 : * on this page.
1251 : */
1252 : Assert(freespace >= sizeof(uint32));
1253 :
1254 : /* Copy record data */
1255 29231760 : written = 0;
1256 138113912 : while (rdata != NULL)
1257 : {
1258 108882152 : const char *rdata_data = rdata->data;
1259 108882152 : int rdata_len = rdata->len;
1260 :
1261 112436788 : while (rdata_len > freespace)
1262 : {
1263 : /*
1264 : * Write what fits on this page, and continue on the next page.
1265 : */
1266 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1267 3554636 : memcpy(currpos, rdata_data, freespace);
1268 3554636 : rdata_data += freespace;
1269 3554636 : rdata_len -= freespace;
1270 3554636 : written += freespace;
1271 3554636 : CurrPos += freespace;
1272 :
1273 : /*
1274 : * Get pointer to beginning of next page, and set the xlp_rem_len
1275 : * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1276 : *
1277 : * It's safe to set the contrecord flag and xlp_rem_len without a
1278 : * lock on the page. All the other flags were already set when the
1279 : * page was initialized, in AdvanceXLInsertBuffer, and we're the
1280 : * only backend that needs to set the contrecord flag.
1281 : */
1282 3554636 : currpos = GetXLogBuffer(CurrPos, tli);
1283 3554636 : pagehdr = (XLogPageHeader) currpos;
1284 3554636 : pagehdr->xlp_rem_len = write_len - written;
1285 3554636 : pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1286 :
1287 : /* skip over the page header */
1288 3554636 : if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1289 : {
1290 2362 : CurrPos += SizeOfXLogLongPHD;
1291 2362 : currpos += SizeOfXLogLongPHD;
1292 : }
1293 : else
1294 : {
1295 3552274 : CurrPos += SizeOfXLogShortPHD;
1296 3552274 : currpos += SizeOfXLogShortPHD;
1297 : }
1298 3554636 : freespace = INSERT_FREESPACE(CurrPos);
1299 : }
1300 :
1301 : Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1302 108882152 : memcpy(currpos, rdata_data, rdata_len);
1303 108882152 : currpos += rdata_len;
1304 108882152 : CurrPos += rdata_len;
1305 108882152 : freespace -= rdata_len;
1306 108882152 : written += rdata_len;
1307 :
1308 108882152 : rdata = rdata->next;
1309 : }
1310 : Assert(written == write_len);
1311 :
1312 : /*
1313 : * If this was an xlog-switch, it's not enough to write the switch record,
1314 : * we also have to consume all the remaining space in the WAL segment. We
1315 : * have already reserved that space, but we need to actually fill it.
1316 : */
1317 29231760 : if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1318 : {
1319 : /* An xlog-switch record doesn't contain any data besides the header */
1320 : Assert(write_len == SizeOfXLogRecord);
1321 :
1322 : /* Assert that we did reserve the right amount of space */
1323 : Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1324 :
1325 : /* Use up all the remaining space on the current page */
1326 1450 : CurrPos += freespace;
1327 :
1328 : /*
1329 : * Cause all remaining pages in the segment to be flushed, leaving the
1330 : * XLog position where it should be, at the start of the next segment.
1331 : * We do this one page at a time, to make sure we don't deadlock
1332 : * against ourselves if wal_buffers < wal_segment_size.
1333 : */
1334 1397926 : while (CurrPos < EndPos)
1335 : {
1336 : /*
1337 : * The minimal action to flush the page would be to call
1338 : * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1339 : * AdvanceXLInsertBuffer(...). The page would be left initialized
1340 : * mostly to zeros, except for the page header (always the short
1341 : * variant, as this is never a segment's first page).
1342 : *
1343 : * The large vistas of zeros are good for compressibility, but the
1344 : * headers interrupting them every XLOG_BLCKSZ (with values that
1345 : * differ from page to page) are not. The effect varies with
1346 : * compression tool, but bzip2 for instance compresses about an
1347 : * order of magnitude worse if those headers are left in place.
1348 : *
1349 : * Rather than complicating AdvanceXLInsertBuffer itself (which is
1350 : * called in heavily-loaded circumstances as well as this lightly-
1351 : * loaded one) with variant behavior, we just use GetXLogBuffer
1352 : * (which itself calls the two methods we need) to get the pointer
1353 : * and zero most of the page. Then we just zero the page header.
1354 : */
1355 1396476 : currpos = GetXLogBuffer(CurrPos, tli);
1356 5585904 : MemSet(currpos, 0, SizeOfXLogShortPHD);
1357 :
1358 1396476 : CurrPos += XLOG_BLCKSZ;
1359 : }
1360 : }
1361 : else
1362 : {
1363 : /* Align the end position, so that the next record starts aligned */
1364 29230310 : CurrPos = MAXALIGN64(CurrPos);
1365 : }
1366 :
1367 29231760 : if (CurrPos != EndPos)
1368 0 : ereport(PANIC,
1369 : errcode(ERRCODE_DATA_CORRUPTED),
1370 : errmsg_internal("space reserved for WAL record does not match what was written"));
1371 29231760 : }
1372 :
1373 : /*
1374 : * Acquire a WAL insertion lock, for inserting to WAL.
1375 : */
1376 : static void
1377 29243874 : WALInsertLockAcquire(void)
1378 : {
1379 : bool immed;
1380 :
1381 : /*
1382 : * It doesn't matter which of the WAL insertion locks we acquire, so try
1383 : * the one we used last time. If the system isn't particularly busy, it's
1384 : * a good bet that it's still available, and it's good to have some
1385 : * affinity to a particular lock so that you don't unnecessarily bounce
1386 : * cache lines between processes when there's no contention.
1387 : *
1388 : * If this is the first time through in this backend, pick a lock
1389 : * (semi-)randomly. This allows the locks to be used evenly if you have a
1390 : * lot of very short connections.
1391 : */
1392 : static int lockToTry = -1;
1393 :
1394 29243874 : if (lockToTry == -1)
1395 16132 : lockToTry = MyProcNumber % NUM_XLOGINSERT_LOCKS;
1396 29243874 : MyLockNo = lockToTry;
1397 :
1398 : /*
1399 : * The insertingAt value is initially set to 0, as we don't know our
1400 : * insert location yet.
1401 : */
1402 29243874 : immed = LWLockAcquire(&WALInsertLocks[MyLockNo].l.lock, LW_EXCLUSIVE);
1403 29243874 : if (!immed)
1404 : {
1405 : /*
1406 : * If we couldn't get the lock immediately, try another lock next
1407 : * time. On a system with more insertion locks than concurrent
1408 : * inserters, this causes all the inserters to eventually migrate to a
1409 : * lock that no-one else is using. On a system with more inserters
1410 : * than locks, it still helps to distribute the inserters evenly
1411 : * across the locks.
1412 : */
1413 72002 : lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1414 : }
1415 29243874 : }
1416 :
1417 : /*
1418 : * Acquire all WAL insertion locks, to prevent other backends from inserting
1419 : * to WAL.
1420 : */
1421 : static void
1422 8664 : WALInsertLockAcquireExclusive(void)
1423 : {
1424 : int i;
1425 :
1426 : /*
1427 : * When holding all the locks, all but the last lock's insertingAt
1428 : * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1429 : * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1430 : */
1431 69312 : for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1432 : {
1433 60648 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1434 60648 : LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1435 60648 : &WALInsertLocks[i].l.insertingAt,
1436 : PG_UINT64_MAX);
1437 : }
1438 : /* Variable value reset to 0 at release */
1439 8664 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
1440 :
1441 8664 : holdingAllLocks = true;
1442 8664 : }
1443 :
1444 : /*
1445 : * Release our insertion lock (or locks, if we're holding them all).
1446 : *
1447 : * NB: Reset all variables to 0, so they cause LWLockWaitForVar to block the
1448 : * next time the lock is acquired.
1449 : */
1450 : static void
1451 29252538 : WALInsertLockRelease(void)
1452 : {
1453 29252538 : if (holdingAllLocks)
1454 : {
1455 : int i;
1456 :
1457 77976 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1458 69312 : LWLockReleaseClearVar(&WALInsertLocks[i].l.lock,
1459 69312 : &WALInsertLocks[i].l.insertingAt,
1460 : 0);
1461 :
1462 8664 : holdingAllLocks = false;
1463 : }
1464 : else
1465 : {
1466 29243874 : LWLockReleaseClearVar(&WALInsertLocks[MyLockNo].l.lock,
1467 29243874 : &WALInsertLocks[MyLockNo].l.insertingAt,
1468 : 0);
1469 : }
1470 29252538 : }
1471 :
1472 : /*
1473 : * Update our insertingAt value, to let others know that we've finished
1474 : * inserting up to that point.
1475 : */
1476 : static void
1477 5396158 : WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
1478 : {
1479 5396158 : if (holdingAllLocks)
1480 : {
1481 : /*
1482 : * We use the last lock to mark our actual position, see comments in
1483 : * WALInsertLockAcquireExclusive.
1484 : */
1485 1390150 : LWLockUpdateVar(&WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.lock,
1486 1390150 : &WALInsertLocks[NUM_XLOGINSERT_LOCKS - 1].l.insertingAt,
1487 : insertingAt);
1488 : }
1489 : else
1490 4006008 : LWLockUpdateVar(&WALInsertLocks[MyLockNo].l.lock,
1491 4006008 : &WALInsertLocks[MyLockNo].l.insertingAt,
1492 : insertingAt);
1493 5396158 : }
1494 :
1495 : /*
1496 : * Wait for any WAL insertions < upto to finish.
1497 : *
1498 : * Returns the location of the oldest insertion that is still in-progress.
1499 : * Any WAL prior to that point has been fully copied into WAL buffers, and
1500 : * can be flushed out to disk. Because this waits for any insertions older
1501 : * than 'upto' to finish, the return value is always >= 'upto'.
1502 : *
1503 : * Note: When you are about to write out WAL, you must call this function
1504 : * *before* acquiring WALWriteLock, to avoid deadlocks. This function might
1505 : * need to wait for an insertion to finish (or at least advance to next
1506 : * uninitialized page), and the inserter might need to evict an old WAL buffer
1507 : * to make room for a new one, which in turn requires WALWriteLock.
1508 : */
1509 : static XLogRecPtr
1510 4920016 : WaitXLogInsertionsToFinish(XLogRecPtr upto)
1511 : {
1512 : uint64 bytepos;
1513 : XLogRecPtr inserted;
1514 : XLogRecPtr reservedUpto;
1515 : XLogRecPtr finishedUpto;
1516 4920016 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1517 : int i;
1518 :
1519 4920016 : if (MyProc == NULL)
1520 0 : elog(PANIC, "cannot wait without a PGPROC structure");
1521 :
1522 : /*
1523 : * Check if there's any work to do. Use a barrier to ensure we get the
1524 : * freshest value.
1525 : */
1526 4920016 : inserted = pg_atomic_read_membarrier_u64(&XLogCtl->logInsertResult);
1527 4920016 : if (upto <= inserted)
1528 4066978 : return inserted;
1529 :
1530 : /* Read the current insert position */
1531 853038 : SpinLockAcquire(&Insert->insertpos_lck);
1532 853038 : bytepos = Insert->CurrBytePos;
1533 853038 : SpinLockRelease(&Insert->insertpos_lck);
1534 853038 : reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1535 :
1536 : /*
1537 : * No-one should request to flush a piece of WAL that hasn't even been
1538 : * reserved yet. However, it can happen if there is a block with a bogus
1539 : * LSN on disk, for example. XLogFlush checks for that situation and
1540 : * complains, but only after the flush. Here we just assume that to mean
1541 : * that all WAL that has been reserved needs to be finished. In this
1542 : * corner-case, the return value can be smaller than 'upto' argument.
1543 : */
1544 853038 : if (upto > reservedUpto)
1545 : {
1546 0 : ereport(LOG,
1547 : errmsg("request to flush past end of generated WAL; request %X/%08X, current position %X/%08X",
1548 : LSN_FORMAT_ARGS(upto), LSN_FORMAT_ARGS(reservedUpto)));
1549 0 : upto = reservedUpto;
1550 : }
1551 :
1552 : /*
1553 : * Loop through all the locks, sleeping on any in-progress insert older
1554 : * than 'upto'.
1555 : *
1556 : * finishedUpto is our return value, indicating the point upto which all
1557 : * the WAL insertions have been finished. Initialize it to the head of
1558 : * reserved WAL, and as we iterate through the insertion locks, back it
1559 : * out for any insertion that's still in progress.
1560 : */
1561 853038 : finishedUpto = reservedUpto;
1562 7677342 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1563 : {
1564 6824304 : XLogRecPtr insertingat = InvalidXLogRecPtr;
1565 :
1566 : do
1567 : {
1568 : /*
1569 : * See if this insertion is in progress. LWLockWaitForVar will
1570 : * wait for the lock to be released, or for the 'value' to be set
1571 : * by a LWLockUpdateVar call. When a lock is initially acquired,
1572 : * its value is 0 (InvalidXLogRecPtr), which means that we don't
1573 : * know where it's inserting yet. We will have to wait for it. If
1574 : * it's a small insertion, the record will most likely fit on the
1575 : * same page and the inserter will release the lock without ever
1576 : * calling LWLockUpdateVar. But if it has to sleep, it will
1577 : * advertise the insertion point with LWLockUpdateVar before
1578 : * sleeping.
1579 : *
1580 : * In this loop we are only waiting for insertions that started
1581 : * before WaitXLogInsertionsToFinish was called. The lack of
1582 : * memory barriers in the loop means that we might see locks as
1583 : * "unused" that have since become used. This is fine because
1584 : * they only can be used for later insertions that we would not
1585 : * want to wait on anyway. Not taking a lock to acquire the
1586 : * current insertingAt value means that we might see older
1587 : * insertingAt values. This is also fine, because if we read a
1588 : * value too old, we will add ourselves to the wait queue, which
1589 : * contains atomic operations.
1590 : */
1591 6954718 : if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1592 6954718 : &WALInsertLocks[i].l.insertingAt,
1593 : insertingat, &insertingat))
1594 : {
1595 : /* the lock was free, so no insertion in progress */
1596 4949776 : insertingat = InvalidXLogRecPtr;
1597 4949776 : break;
1598 : }
1599 :
1600 : /*
1601 : * This insertion is still in progress. Have to wait, unless the
1602 : * inserter has proceeded past 'upto'.
1603 : */
1604 2004942 : } while (insertingat < upto);
1605 :
1606 6824304 : if (XLogRecPtrIsValid(insertingat) && insertingat < finishedUpto)
1607 739726 : finishedUpto = insertingat;
1608 : }
1609 :
1610 : /*
1611 : * Advance the limit we know to have been inserted and return the freshest
1612 : * value we know of, which might be beyond what we requested if somebody
1613 : * is concurrently doing this with an 'upto' pointer ahead of us.
1614 : */
1615 853038 : finishedUpto = pg_atomic_monotonic_advance_u64(&XLogCtl->logInsertResult,
1616 : finishedUpto);
1617 :
1618 853038 : return finishedUpto;
1619 : }
1620 :
1621 : /*
1622 : * Get a pointer to the right location in the WAL buffer containing the
1623 : * given XLogRecPtr.
1624 : *
1625 : * If the page is not initialized yet, it is initialized. That might require
1626 : * evicting an old dirty buffer from the buffer cache, which means I/O.
1627 : *
1628 : * The caller must ensure that the page containing the requested location
1629 : * isn't evicted yet, and won't be evicted. The way to ensure that is to
1630 : * hold onto a WAL insertion lock with the insertingAt position set to
1631 : * something <= ptr. GetXLogBuffer() will update insertingAt if it needs
1632 : * to evict an old page from the buffer. (This means that once you call
1633 : * GetXLogBuffer() with a given 'ptr', you must not access anything before
1634 : * that point anymore, and must not call GetXLogBuffer() with an older 'ptr'
1635 : * later, because older buffers might be recycled already)
1636 : */
1637 : static char *
1638 34182894 : GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli)
1639 : {
1640 : int idx;
1641 : XLogRecPtr endptr;
1642 : static uint64 cachedPage = 0;
1643 : static char *cachedPos = NULL;
1644 : XLogRecPtr expectedEndPtr;
1645 :
1646 : /*
1647 : * Fast path for the common case that we need to access again the same
1648 : * page as last time.
1649 : */
1650 34182894 : if (ptr / XLOG_BLCKSZ == cachedPage)
1651 : {
1652 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1653 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1654 27683802 : return cachedPos + ptr % XLOG_BLCKSZ;
1655 : }
1656 :
1657 : /*
1658 : * The XLog buffer cache is organized so that a page is always loaded to a
1659 : * particular buffer. That way we can easily calculate the buffer a given
1660 : * page must be loaded into, from the XLogRecPtr alone.
1661 : */
1662 6499092 : idx = XLogRecPtrToBufIdx(ptr);
1663 :
1664 : /*
1665 : * See what page is loaded in the buffer at the moment. It could be the
1666 : * page we're looking for, or something older. It can't be anything newer
1667 : * - that would imply the page we're looking for has already been written
1668 : * out to disk and evicted, and the caller is responsible for making sure
1669 : * that doesn't happen.
1670 : *
1671 : * We don't hold a lock while we read the value. If someone is just about
1672 : * to initialize or has just initialized the page, it's possible that we
1673 : * get InvalidXLogRecPtr. That's ok, we'll grab the mapping lock (in
1674 : * AdvanceXLInsertBuffer) and retry if we see anything other than the page
1675 : * we're looking for.
1676 : */
1677 6499092 : expectedEndPtr = ptr;
1678 6499092 : expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1679 :
1680 6499092 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1681 6499092 : if (expectedEndPtr != endptr)
1682 : {
1683 : XLogRecPtr initializedUpto;
1684 :
1685 : /*
1686 : * Before calling AdvanceXLInsertBuffer(), which can block, let others
1687 : * know how far we're finished with inserting the record.
1688 : *
1689 : * NB: If 'ptr' points to just after the page header, advertise a
1690 : * position at the beginning of the page rather than 'ptr' itself. If
1691 : * there are no other insertions running, someone might try to flush
1692 : * up to our advertised location. If we advertised a position after
1693 : * the page header, someone might try to flush the page header, even
1694 : * though page might actually not be initialized yet. As the first
1695 : * inserter on the page, we are effectively responsible for making
1696 : * sure that it's initialized, before we let insertingAt to move past
1697 : * the page header.
1698 : */
1699 5396158 : if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
1700 12360 : XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
1701 12360 : initializedUpto = ptr - SizeOfXLogShortPHD;
1702 5383798 : else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
1703 2714 : XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
1704 1224 : initializedUpto = ptr - SizeOfXLogLongPHD;
1705 : else
1706 5382574 : initializedUpto = ptr;
1707 :
1708 5396158 : WALInsertLockUpdateInsertingAt(initializedUpto);
1709 :
1710 5396158 : AdvanceXLInsertBuffer(ptr, tli, false);
1711 5396158 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1712 :
1713 5396158 : if (expectedEndPtr != endptr)
1714 0 : elog(PANIC, "could not find WAL buffer for %X/%08X",
1715 : LSN_FORMAT_ARGS(ptr));
1716 : }
1717 : else
1718 : {
1719 : /*
1720 : * Make sure the initialization of the page is visible to us, and
1721 : * won't arrive later to overwrite the WAL data we write on the page.
1722 : */
1723 1102934 : pg_memory_barrier();
1724 : }
1725 :
1726 : /*
1727 : * Found the buffer holding this page. Return a pointer to the right
1728 : * offset within the page.
1729 : */
1730 6499092 : cachedPage = ptr / XLOG_BLCKSZ;
1731 6499092 : cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1732 :
1733 : Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1734 : Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1735 :
1736 6499092 : return cachedPos + ptr % XLOG_BLCKSZ;
1737 : }
1738 :
1739 : /*
1740 : * Read WAL data directly from WAL buffers, if available. Returns the number
1741 : * of bytes read successfully.
1742 : *
1743 : * Fewer than 'count' bytes may be read if some of the requested WAL data has
1744 : * already been evicted.
1745 : *
1746 : * No locks are taken.
1747 : *
1748 : * Caller should ensure that it reads no further than LogwrtResult.Write
1749 : * (which should have been updated by the caller when determining how far to
1750 : * read). The 'tli' argument is only used as a convenient safety check so that
1751 : * callers do not read from WAL buffers on a historical timeline.
1752 : */
1753 : Size
1754 198312 : WALReadFromBuffers(char *dstbuf, XLogRecPtr startptr, Size count,
1755 : TimeLineID tli)
1756 : {
1757 198312 : char *pdst = dstbuf;
1758 198312 : XLogRecPtr recptr = startptr;
1759 : XLogRecPtr inserted;
1760 198312 : Size nbytes = count;
1761 :
1762 198312 : if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1763 1502 : return 0;
1764 :
1765 : Assert(XLogRecPtrIsValid(startptr));
1766 :
1767 : /*
1768 : * Caller should ensure that the requested data has been inserted into WAL
1769 : * buffers before we try to read it.
1770 : */
1771 196810 : inserted = pg_atomic_read_u64(&XLogCtl->logInsertResult);
1772 196810 : if (startptr + count > inserted)
1773 0 : ereport(ERROR,
1774 : errmsg("cannot read past end of generated WAL: requested %X/%08X, current position %X/%08X",
1775 : LSN_FORMAT_ARGS(startptr + count),
1776 : LSN_FORMAT_ARGS(inserted)));
1777 :
1778 : /*
1779 : * Loop through the buffers without a lock. For each buffer, atomically
1780 : * read and verify the end pointer, then copy the data out, and finally
1781 : * re-read and re-verify the end pointer.
1782 : *
1783 : * Once a page is evicted, it never returns to the WAL buffers, so if the
1784 : * end pointer matches the expected end pointer before and after we copy
1785 : * the data, then the right page must have been present during the data
1786 : * copy. Read barriers are necessary to ensure that the data copy actually
1787 : * happens between the two verification steps.
1788 : *
1789 : * If either verification fails, we simply terminate the loop and return
1790 : * with the data that had been already copied out successfully.
1791 : */
1792 209746 : while (nbytes > 0)
1793 : {
1794 206570 : uint32 offset = recptr % XLOG_BLCKSZ;
1795 206570 : int idx = XLogRecPtrToBufIdx(recptr);
1796 : XLogRecPtr expectedEndPtr;
1797 : XLogRecPtr endptr;
1798 : const char *page;
1799 : const char *psrc;
1800 : Size npagebytes;
1801 :
1802 : /*
1803 : * Calculate the end pointer we expect in the xlblocks array if the
1804 : * correct page is present.
1805 : */
1806 206570 : expectedEndPtr = recptr + (XLOG_BLCKSZ - offset);
1807 :
1808 : /*
1809 : * First verification step: check that the correct page is present in
1810 : * the WAL buffers.
1811 : */
1812 206570 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1813 206570 : if (expectedEndPtr != endptr)
1814 193630 : break;
1815 :
1816 : /*
1817 : * The correct page is present (or was at the time the endptr was
1818 : * read; must re-verify later). Calculate pointer to source data and
1819 : * determine how much data to read from this page.
1820 : */
1821 12940 : page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1822 12940 : psrc = page + offset;
1823 12940 : npagebytes = Min(nbytes, XLOG_BLCKSZ - offset);
1824 :
1825 : /*
1826 : * Ensure that the data copy and the first verification step are not
1827 : * reordered.
1828 : */
1829 12940 : pg_read_barrier();
1830 :
1831 : /* data copy */
1832 12940 : memcpy(pdst, psrc, npagebytes);
1833 :
1834 : /*
1835 : * Ensure that the data copy and the second verification step are not
1836 : * reordered.
1837 : */
1838 12940 : pg_read_barrier();
1839 :
1840 : /*
1841 : * Second verification step: check that the page we read from wasn't
1842 : * evicted while we were copying the data.
1843 : */
1844 12940 : endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1845 12940 : if (expectedEndPtr != endptr)
1846 4 : break;
1847 :
1848 12936 : pdst += npagebytes;
1849 12936 : recptr += npagebytes;
1850 12936 : nbytes -= npagebytes;
1851 : }
1852 :
1853 : Assert(pdst - dstbuf <= count);
1854 :
1855 196810 : return pdst - dstbuf;
1856 : }
1857 :
1858 : /*
1859 : * Converts a "usable byte position" to XLogRecPtr. A usable byte position
1860 : * is the position starting from the beginning of WAL, excluding all WAL
1861 : * page headers.
1862 : */
1863 : static XLogRecPtr
1864 58468770 : XLogBytePosToRecPtr(uint64 bytepos)
1865 : {
1866 : uint64 fullsegs;
1867 : uint64 fullpages;
1868 : uint64 bytesleft;
1869 : uint32 seg_offset;
1870 : XLogRecPtr result;
1871 :
1872 58468770 : fullsegs = bytepos / UsableBytesInSegment;
1873 58468770 : bytesleft = bytepos % UsableBytesInSegment;
1874 :
1875 58468770 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1876 : {
1877 : /* fits on first page of segment */
1878 102868 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1879 : }
1880 : else
1881 : {
1882 : /* account for the first page on segment with long header */
1883 58365902 : seg_offset = XLOG_BLCKSZ;
1884 58365902 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1885 :
1886 58365902 : fullpages = bytesleft / UsableBytesInPage;
1887 58365902 : bytesleft = bytesleft % UsableBytesInPage;
1888 :
1889 58365902 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1890 : }
1891 :
1892 58468770 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1893 :
1894 58468770 : return result;
1895 : }
1896 :
1897 : /*
1898 : * Like XLogBytePosToRecPtr, but if the position is at a page boundary,
1899 : * returns a pointer to the beginning of the page (ie. before page header),
1900 : * not to where the first xlog record on that page would go to. This is used
1901 : * when converting a pointer to the end of a record.
1902 : */
1903 : static XLogRecPtr
1904 30086364 : XLogBytePosToEndRecPtr(uint64 bytepos)
1905 : {
1906 : uint64 fullsegs;
1907 : uint64 fullpages;
1908 : uint64 bytesleft;
1909 : uint32 seg_offset;
1910 : XLogRecPtr result;
1911 :
1912 30086364 : fullsegs = bytepos / UsableBytesInSegment;
1913 30086364 : bytesleft = bytepos % UsableBytesInSegment;
1914 :
1915 30086364 : if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1916 : {
1917 : /* fits on first page of segment */
1918 165144 : if (bytesleft == 0)
1919 111088 : seg_offset = 0;
1920 : else
1921 54056 : seg_offset = bytesleft + SizeOfXLogLongPHD;
1922 : }
1923 : else
1924 : {
1925 : /* account for the first page on segment with long header */
1926 29921220 : seg_offset = XLOG_BLCKSZ;
1927 29921220 : bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1928 :
1929 29921220 : fullpages = bytesleft / UsableBytesInPage;
1930 29921220 : bytesleft = bytesleft % UsableBytesInPage;
1931 :
1932 29921220 : if (bytesleft == 0)
1933 28502 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1934 : else
1935 29892718 : seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1936 : }
1937 :
1938 30086364 : XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1939 :
1940 30086364 : return result;
1941 : }
1942 :
1943 : /*
1944 : * Convert an XLogRecPtr to a "usable byte position".
1945 : */
1946 : static uint64
1947 5042 : XLogRecPtrToBytePos(XLogRecPtr ptr)
1948 : {
1949 : uint64 fullsegs;
1950 : uint32 fullpages;
1951 : uint32 offset;
1952 : uint64 result;
1953 :
1954 5042 : XLByteToSeg(ptr, fullsegs, wal_segment_size);
1955 :
1956 5042 : fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
1957 5042 : offset = ptr % XLOG_BLCKSZ;
1958 :
1959 5042 : if (fullpages == 0)
1960 : {
1961 1980 : result = fullsegs * UsableBytesInSegment;
1962 1980 : if (offset > 0)
1963 : {
1964 : Assert(offset >= SizeOfXLogLongPHD);
1965 496 : result += offset - SizeOfXLogLongPHD;
1966 : }
1967 : }
1968 : else
1969 : {
1970 3062 : result = fullsegs * UsableBytesInSegment +
1971 3062 : (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
1972 3062 : (fullpages - 1) * UsableBytesInPage; /* full pages */
1973 3062 : if (offset > 0)
1974 : {
1975 : Assert(offset >= SizeOfXLogShortPHD);
1976 3036 : result += offset - SizeOfXLogShortPHD;
1977 : }
1978 : }
1979 :
1980 5042 : return result;
1981 : }
1982 :
1983 : /*
1984 : * Initialize XLOG buffers, writing out old buffers if they still contain
1985 : * unwritten data, upto the page containing 'upto'. Or if 'opportunistic' is
1986 : * true, initialize as many pages as we can without having to write out
1987 : * unwritten data. Any new pages are initialized to zeros, with pages headers
1988 : * initialized properly.
1989 : */
1990 : static void
1991 5405128 : AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
1992 : {
1993 5405128 : XLogCtlInsert *Insert = &XLogCtl->Insert;
1994 : int nextidx;
1995 : XLogRecPtr OldPageRqstPtr;
1996 : XLogwrtRqst WriteRqst;
1997 5405128 : XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
1998 : XLogRecPtr NewPageBeginPtr;
1999 : XLogPageHeader NewPage;
2000 5405128 : int npages pg_attribute_unused() = 0;
2001 :
2002 5405128 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2003 :
2004 : /*
2005 : * Now that we have the lock, check if someone initialized the page
2006 : * already.
2007 : */
2008 15107190 : while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2009 : {
2010 9711032 : nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
2011 :
2012 : /*
2013 : * Get ending-offset of the buffer page we need to replace (this may
2014 : * be zero if the buffer hasn't been used yet). Fall through if it's
2015 : * already written out.
2016 : */
2017 9711032 : OldPageRqstPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[nextidx]);
2018 9711032 : if (LogwrtResult.Write < OldPageRqstPtr)
2019 : {
2020 : /*
2021 : * Nope, got work to do. If we just want to pre-initialize as much
2022 : * as we can without flushing, give up now.
2023 : */
2024 4816954 : if (opportunistic)
2025 8970 : break;
2026 :
2027 : /* Advance shared memory write request position */
2028 4807984 : SpinLockAcquire(&XLogCtl->info_lck);
2029 4807984 : if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2030 1259294 : XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2031 4807984 : SpinLockRelease(&XLogCtl->info_lck);
2032 :
2033 : /*
2034 : * Acquire an up-to-date LogwrtResult value and see if we still
2035 : * need to write it or if someone else already did.
2036 : */
2037 4807984 : RefreshXLogWriteResult(LogwrtResult);
2038 4807984 : if (LogwrtResult.Write < OldPageRqstPtr)
2039 : {
2040 : /*
2041 : * Must acquire write lock. Release WALBufMappingLock first,
2042 : * to make sure that all insertions that we need to wait for
2043 : * can finish (up to this same position). Otherwise we risk
2044 : * deadlock.
2045 : */
2046 4641680 : LWLockRelease(WALBufMappingLock);
2047 :
2048 4641680 : WaitXLogInsertionsToFinish(OldPageRqstPtr);
2049 :
2050 4641680 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2051 :
2052 4641680 : RefreshXLogWriteResult(LogwrtResult);
2053 4641680 : if (LogwrtResult.Write >= OldPageRqstPtr)
2054 : {
2055 : /* OK, someone wrote it already */
2056 683774 : LWLockRelease(WALWriteLock);
2057 : }
2058 : else
2059 : {
2060 : /* Have to write it ourselves */
2061 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2062 3957906 : WriteRqst.Write = OldPageRqstPtr;
2063 3957906 : WriteRqst.Flush = 0;
2064 3957906 : XLogWrite(WriteRqst, tli, false);
2065 3957906 : LWLockRelease(WALWriteLock);
2066 3957906 : pgWalUsage.wal_buffers_full++;
2067 : TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2068 :
2069 : /*
2070 : * Required for the flush of pending stats WAL data, per
2071 : * update of pgWalUsage.
2072 : */
2073 3957906 : pgstat_report_fixed = true;
2074 : }
2075 : /* Re-acquire WALBufMappingLock and retry */
2076 4641680 : LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2077 4641680 : continue;
2078 : }
2079 : }
2080 :
2081 : /*
2082 : * Now the next buffer slot is free and we can set it up to be the
2083 : * next output page.
2084 : */
2085 5060382 : NewPageBeginPtr = XLogCtl->InitializedUpTo;
2086 5060382 : NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2087 :
2088 : Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2089 :
2090 5060382 : NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2091 :
2092 : /*
2093 : * Mark the xlblock with InvalidXLogRecPtr and issue a write barrier
2094 : * before initializing. Otherwise, the old page may be partially
2095 : * zeroed but look valid.
2096 : */
2097 5060382 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], InvalidXLogRecPtr);
2098 5060382 : pg_write_barrier();
2099 :
2100 : /*
2101 : * Be sure to re-zero the buffer so that bytes beyond what we've
2102 : * written will look like zeroes and not valid XLOG records...
2103 : */
2104 5060382 : MemSet(NewPage, 0, XLOG_BLCKSZ);
2105 :
2106 : /*
2107 : * Fill the new page's header
2108 : */
2109 5060382 : NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2110 :
2111 : /* NewPage->xlp_info = 0; */ /* done by memset */
2112 5060382 : NewPage->xlp_tli = tli;
2113 5060382 : NewPage->xlp_pageaddr = NewPageBeginPtr;
2114 :
2115 : /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2116 :
2117 : /*
2118 : * If online backup is not in progress, mark the header to indicate
2119 : * that WAL records beginning in this page have removable backup
2120 : * blocks. This allows the WAL archiver to know whether it is safe to
2121 : * compress archived WAL data by transforming full-block records into
2122 : * the non-full-block format. It is sufficient to record this at the
2123 : * page level because we force a page switch (in fact a segment
2124 : * switch) when starting a backup, so the flag will be off before any
2125 : * records can be written during the backup. At the end of a backup,
2126 : * the last page will be marked as all unsafe when perhaps only part
2127 : * is unsafe, but at worst the archiver would miss the opportunity to
2128 : * compress a few records.
2129 : */
2130 5060382 : if (Insert->runningBackups == 0)
2131 4803046 : NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2132 :
2133 : /*
2134 : * If first page of an XLOG segment file, make it a long header.
2135 : */
2136 5060382 : if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2137 : {
2138 3624 : XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2139 :
2140 3624 : NewLongPage->xlp_sysid = ControlFile->system_identifier;
2141 3624 : NewLongPage->xlp_seg_size = wal_segment_size;
2142 3624 : NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
2143 3624 : NewPage->xlp_info |= XLP_LONG_HEADER;
2144 : }
2145 :
2146 : /*
2147 : * Make sure the initialization of the page becomes visible to others
2148 : * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2149 : * holding a lock.
2150 : */
2151 5060382 : pg_write_barrier();
2152 :
2153 5060382 : pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], NewPageEndPtr);
2154 5060382 : XLogCtl->InitializedUpTo = NewPageEndPtr;
2155 :
2156 5060382 : npages++;
2157 : }
2158 5405128 : LWLockRelease(WALBufMappingLock);
2159 :
2160 : #ifdef WAL_DEBUG
2161 : if (XLOG_DEBUG && npages > 0)
2162 : {
2163 : elog(DEBUG1, "initialized %d pages, up to %X/%08X",
2164 : npages, LSN_FORMAT_ARGS(NewPageEndPtr));
2165 : }
2166 : #endif
2167 5405128 : }
2168 :
2169 : /*
2170 : * Calculate CheckPointSegments based on max_wal_size_mb and
2171 : * checkpoint_completion_target.
2172 : */
2173 : static void
2174 15466 : CalculateCheckpointSegments(void)
2175 : {
2176 : double target;
2177 :
2178 : /*-------
2179 : * Calculate the distance at which to trigger a checkpoint, to avoid
2180 : * exceeding max_wal_size_mb. This is based on two assumptions:
2181 : *
2182 : * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2183 : * WAL for two checkpoint cycles to allow us to recover from the
2184 : * secondary checkpoint if the first checkpoint failed, though we
2185 : * only did this on the primary anyway, not on standby. Keeping just
2186 : * one checkpoint simplifies processing and reduces disk space in
2187 : * many smaller databases.)
2188 : * b) during checkpoint, we consume checkpoint_completion_target *
2189 : * number of segments consumed between checkpoints.
2190 : *-------
2191 : */
2192 15466 : target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2193 15466 : (1.0 + CheckPointCompletionTarget);
2194 :
2195 : /* round down */
2196 15466 : CheckPointSegments = (int) target;
2197 :
2198 15466 : if (CheckPointSegments < 1)
2199 20 : CheckPointSegments = 1;
2200 15466 : }
2201 :
2202 : void
2203 11180 : assign_max_wal_size(int newval, void *extra)
2204 : {
2205 11180 : max_wal_size_mb = newval;
2206 11180 : CalculateCheckpointSegments();
2207 11180 : }
2208 :
2209 : void
2210 2272 : assign_checkpoint_completion_target(double newval, void *extra)
2211 : {
2212 2272 : CheckPointCompletionTarget = newval;
2213 2272 : CalculateCheckpointSegments();
2214 2272 : }
2215 :
2216 : bool
2217 4388 : check_wal_segment_size(int *newval, void **extra, GucSource source)
2218 : {
2219 4388 : if (!IsValidWalSegSize(*newval))
2220 : {
2221 0 : GUC_check_errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
2222 0 : return false;
2223 : }
2224 :
2225 4388 : return true;
2226 : }
2227 :
2228 : /*
2229 : * At a checkpoint, how many WAL segments to recycle as preallocated future
2230 : * XLOG segments? Returns the highest segment that should be preallocated.
2231 : */
2232 : static XLogSegNo
2233 3458 : XLOGfileslop(XLogRecPtr lastredoptr)
2234 : {
2235 : XLogSegNo minSegNo;
2236 : XLogSegNo maxSegNo;
2237 : double distance;
2238 : XLogSegNo recycleSegNo;
2239 :
2240 : /*
2241 : * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2242 : * correspond to. Always recycle enough segments to meet the minimum, and
2243 : * remove enough segments to stay below the maximum.
2244 : */
2245 3458 : minSegNo = lastredoptr / wal_segment_size +
2246 3458 : ConvertToXSegs(min_wal_size_mb, wal_segment_size) - 1;
2247 3458 : maxSegNo = lastredoptr / wal_segment_size +
2248 3458 : ConvertToXSegs(max_wal_size_mb, wal_segment_size) - 1;
2249 :
2250 : /*
2251 : * Between those limits, recycle enough segments to get us through to the
2252 : * estimated end of next checkpoint.
2253 : *
2254 : * To estimate where the next checkpoint will finish, assume that the
2255 : * system runs steadily consuming CheckPointDistanceEstimate bytes between
2256 : * every checkpoint.
2257 : */
2258 3458 : distance = (1.0 + CheckPointCompletionTarget) * CheckPointDistanceEstimate;
2259 : /* add 10% for good measure. */
2260 3458 : distance *= 1.10;
2261 :
2262 3458 : recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2263 : wal_segment_size);
2264 :
2265 3458 : if (recycleSegNo < minSegNo)
2266 2452 : recycleSegNo = minSegNo;
2267 3458 : if (recycleSegNo > maxSegNo)
2268 764 : recycleSegNo = maxSegNo;
2269 :
2270 3458 : return recycleSegNo;
2271 : }
2272 :
2273 : /*
2274 : * Check whether we've consumed enough xlog space that a checkpoint is needed.
2275 : *
2276 : * new_segno indicates a log file that has just been filled up (or read
2277 : * during recovery). We measure the distance from RedoRecPtr to new_segno
2278 : * and see if that exceeds CheckPointSegments.
2279 : *
2280 : * Note: it is caller's responsibility that RedoRecPtr is up-to-date.
2281 : */
2282 : bool
2283 10146 : XLogCheckpointNeeded(XLogSegNo new_segno)
2284 : {
2285 : XLogSegNo old_segno;
2286 :
2287 10146 : XLByteToSeg(RedoRecPtr, old_segno, wal_segment_size);
2288 :
2289 10146 : if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2290 6552 : return true;
2291 3594 : return false;
2292 : }
2293 :
2294 : /*
2295 : * Write and/or fsync the log at least as far as WriteRqst indicates.
2296 : *
2297 : * If flexible == true, we don't have to write as far as WriteRqst, but
2298 : * may stop at any convenient boundary (such as a cache or logfile boundary).
2299 : * This option allows us to avoid uselessly issuing multiple writes when a
2300 : * single one would do.
2301 : *
2302 : * Must be called with WALWriteLock held. WaitXLogInsertionsToFinish(WriteRqst)
2303 : * must be called before grabbing the lock, to make sure the data is ready to
2304 : * write.
2305 : */
2306 : static void
2307 4220304 : XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
2308 : {
2309 : bool ispartialpage;
2310 : bool last_iteration;
2311 : bool finishing_seg;
2312 : int curridx;
2313 : int npages;
2314 : int startidx;
2315 : uint32 startoffset;
2316 :
2317 : /* We should always be inside a critical section here */
2318 : Assert(CritSectionCount > 0);
2319 :
2320 : /*
2321 : * Update local LogwrtResult (caller probably did this already, but...)
2322 : */
2323 4220304 : RefreshXLogWriteResult(LogwrtResult);
2324 :
2325 : /*
2326 : * Since successive pages in the xlog cache are consecutively allocated,
2327 : * we can usually gather multiple pages together and issue just one
2328 : * write() call. npages is the number of pages we have determined can be
2329 : * written together; startidx is the cache block index of the first one,
2330 : * and startoffset is the file offset at which it should go. The latter
2331 : * two variables are only valid when npages > 0, but we must initialize
2332 : * all of them to keep the compiler quiet.
2333 : */
2334 4220304 : npages = 0;
2335 4220304 : startidx = 0;
2336 4220304 : startoffset = 0;
2337 :
2338 : /*
2339 : * Within the loop, curridx is the cache block index of the page to
2340 : * consider writing. Begin at the buffer containing the next unwritten
2341 : * page, or last partially written page.
2342 : */
2343 4220304 : curridx = XLogRecPtrToBufIdx(LogwrtResult.Write);
2344 :
2345 9178712 : while (LogwrtResult.Write < WriteRqst.Write)
2346 : {
2347 : /*
2348 : * Make sure we're not ahead of the insert process. This could happen
2349 : * if we're passed a bogus WriteRqst.Write that is past the end of the
2350 : * last page that's been initialized by AdvanceXLInsertBuffer.
2351 : */
2352 5211138 : XLogRecPtr EndPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[curridx]);
2353 :
2354 5211138 : if (LogwrtResult.Write >= EndPtr)
2355 0 : elog(PANIC, "xlog write request %X/%08X is past end of log %X/%08X",
2356 : LSN_FORMAT_ARGS(LogwrtResult.Write),
2357 : LSN_FORMAT_ARGS(EndPtr));
2358 :
2359 : /* Advance LogwrtResult.Write to end of current buffer page */
2360 5211138 : LogwrtResult.Write = EndPtr;
2361 5211138 : ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2362 :
2363 5211138 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2364 : wal_segment_size))
2365 : {
2366 : /*
2367 : * Switch to new logfile segment. We cannot have any pending
2368 : * pages here (since we dump what we have at segment end).
2369 : */
2370 : Assert(npages == 0);
2371 26518 : if (openLogFile >= 0)
2372 11898 : XLogFileClose();
2373 26518 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2374 : wal_segment_size);
2375 26518 : openLogTLI = tli;
2376 :
2377 : /* create/use new log file */
2378 26518 : openLogFile = XLogFileInit(openLogSegNo, tli);
2379 26518 : ReserveExternalFD();
2380 : }
2381 :
2382 : /* Make sure we have the current logfile open */
2383 5211138 : if (openLogFile < 0)
2384 : {
2385 0 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2386 : wal_segment_size);
2387 0 : openLogTLI = tli;
2388 0 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2389 0 : ReserveExternalFD();
2390 : }
2391 :
2392 : /* Add current page to the set of pending pages-to-dump */
2393 5211138 : if (npages == 0)
2394 : {
2395 : /* first of group */
2396 4251216 : startidx = curridx;
2397 4251216 : startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2398 : wal_segment_size);
2399 : }
2400 5211138 : npages++;
2401 :
2402 : /*
2403 : * Dump the set if this will be the last loop iteration, or if we are
2404 : * at the last page of the cache area (since the next page won't be
2405 : * contiguous in memory), or if we are at the end of the logfile
2406 : * segment.
2407 : */
2408 5211138 : last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2409 :
2410 10176764 : finishing_seg = !ispartialpage &&
2411 4965626 : (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2412 :
2413 5211138 : if (last_iteration ||
2414 992742 : curridx == XLogCtl->XLogCacheBlck ||
2415 : finishing_seg)
2416 : {
2417 : char *from;
2418 : Size nbytes;
2419 : Size nleft;
2420 : ssize_t written;
2421 : instr_time start;
2422 :
2423 : /* OK to write the page(s) */
2424 4251216 : from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2425 4251216 : nbytes = npages * (Size) XLOG_BLCKSZ;
2426 4251216 : nleft = nbytes;
2427 : do
2428 : {
2429 4251216 : errno = 0;
2430 :
2431 : /*
2432 : * Measure I/O timing to write WAL data, for pg_stat_io.
2433 : */
2434 4251216 : start = pgstat_prepare_io_time(track_wal_io_timing);
2435 :
2436 4251216 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
2437 4251216 : written = pg_pwrite(openLogFile, from, nleft, startoffset);
2438 4251216 : pgstat_report_wait_end();
2439 :
2440 4251216 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
2441 : IOOP_WRITE, start, 1, written);
2442 :
2443 4251216 : if (written <= 0)
2444 : {
2445 : char xlogfname[MAXFNAMELEN];
2446 : int save_errno;
2447 :
2448 0 : if (errno == EINTR)
2449 0 : continue;
2450 :
2451 0 : save_errno = errno;
2452 0 : XLogFileName(xlogfname, tli, openLogSegNo,
2453 : wal_segment_size);
2454 0 : errno = save_errno;
2455 0 : ereport(PANIC,
2456 : (errcode_for_file_access(),
2457 : errmsg("could not write to log file \"%s\" at offset %u, length %zu: %m",
2458 : xlogfname, startoffset, nleft)));
2459 : }
2460 4251216 : nleft -= written;
2461 4251216 : from += written;
2462 4251216 : startoffset += written;
2463 4251216 : } while (nleft > 0);
2464 :
2465 4251216 : npages = 0;
2466 :
2467 : /*
2468 : * If we just wrote the whole last page of a logfile segment,
2469 : * fsync the segment immediately. This avoids having to go back
2470 : * and re-open prior segments when an fsync request comes along
2471 : * later. Doing it here ensures that one and only one backend will
2472 : * perform this fsync.
2473 : *
2474 : * This is also the right place to notify the Archiver that the
2475 : * segment is ready to copy to archival storage, and to update the
2476 : * timer for archive_timeout, and to signal for a checkpoint if
2477 : * too many logfile segments have been used since the last
2478 : * checkpoint.
2479 : */
2480 4251216 : if (finishing_seg)
2481 : {
2482 3832 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2483 :
2484 : /* signal that we need to wakeup walsenders later */
2485 3832 : WalSndWakeupRequest();
2486 :
2487 3832 : LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2488 :
2489 3832 : if (XLogArchivingActive())
2490 812 : XLogArchiveNotifySeg(openLogSegNo, tli);
2491 :
2492 3832 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
2493 3832 : XLogCtl->lastSegSwitchLSN = LogwrtResult.Flush;
2494 :
2495 : /*
2496 : * Request a checkpoint if we've consumed too much xlog since
2497 : * the last one. For speed, we first check using the local
2498 : * copy of RedoRecPtr, which might be out of date; if it looks
2499 : * like a checkpoint is needed, forcibly update RedoRecPtr and
2500 : * recheck.
2501 : */
2502 3832 : if (IsUnderPostmaster && XLogCheckpointNeeded(openLogSegNo))
2503 : {
2504 496 : (void) GetRedoRecPtr();
2505 496 : if (XLogCheckpointNeeded(openLogSegNo))
2506 396 : RequestCheckpoint(CHECKPOINT_CAUSE_XLOG);
2507 : }
2508 : }
2509 : }
2510 :
2511 5211138 : if (ispartialpage)
2512 : {
2513 : /* Only asked to write a partial page */
2514 245512 : LogwrtResult.Write = WriteRqst.Write;
2515 245512 : break;
2516 : }
2517 4965626 : curridx = NextBufIdx(curridx);
2518 :
2519 : /* If flexible, break out of loop as soon as we wrote something */
2520 4965626 : if (flexible && npages == 0)
2521 7218 : break;
2522 : }
2523 :
2524 : Assert(npages == 0);
2525 :
2526 : /*
2527 : * If asked to flush, do so
2528 : */
2529 4220304 : if (LogwrtResult.Flush < WriteRqst.Flush &&
2530 260918 : LogwrtResult.Flush < LogwrtResult.Write)
2531 : {
2532 : /*
2533 : * Could get here without iterating above loop, in which case we might
2534 : * have no open file or the wrong one. However, we do not need to
2535 : * fsync more than one file.
2536 : */
2537 260784 : if (wal_sync_method != WAL_SYNC_METHOD_OPEN &&
2538 260784 : wal_sync_method != WAL_SYNC_METHOD_OPEN_DSYNC)
2539 : {
2540 260784 : if (openLogFile >= 0 &&
2541 260756 : !XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
2542 : wal_segment_size))
2543 124 : XLogFileClose();
2544 260784 : if (openLogFile < 0)
2545 : {
2546 152 : XLByteToPrevSeg(LogwrtResult.Write, openLogSegNo,
2547 : wal_segment_size);
2548 152 : openLogTLI = tli;
2549 152 : openLogFile = XLogFileOpen(openLogSegNo, tli);
2550 152 : ReserveExternalFD();
2551 : }
2552 :
2553 260784 : issue_xlog_fsync(openLogFile, openLogSegNo, tli);
2554 : }
2555 :
2556 : /* signal that we need to wakeup walsenders later */
2557 260784 : WalSndWakeupRequest();
2558 :
2559 260784 : LogwrtResult.Flush = LogwrtResult.Write;
2560 : }
2561 :
2562 : /*
2563 : * Update shared-memory status
2564 : *
2565 : * We make sure that the shared 'request' values do not fall behind the
2566 : * 'result' values. This is not absolutely essential, but it saves some
2567 : * code in a couple of places.
2568 : */
2569 4220304 : SpinLockAcquire(&XLogCtl->info_lck);
2570 4220304 : if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
2571 226012 : XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
2572 4220304 : if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
2573 263840 : XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
2574 4220304 : SpinLockRelease(&XLogCtl->info_lck);
2575 :
2576 : /*
2577 : * We write Write first, bar, then Flush. When reading, the opposite must
2578 : * be done (with a matching barrier in between), so that we always see a
2579 : * Flush value that trails behind the Write value seen.
2580 : */
2581 4220304 : pg_atomic_write_u64(&XLogCtl->logWriteResult, LogwrtResult.Write);
2582 4220304 : pg_write_barrier();
2583 4220304 : pg_atomic_write_u64(&XLogCtl->logFlushResult, LogwrtResult.Flush);
2584 :
2585 : #ifdef USE_ASSERT_CHECKING
2586 : {
2587 : XLogRecPtr Flush;
2588 : XLogRecPtr Write;
2589 : XLogRecPtr Insert;
2590 :
2591 : Flush = pg_atomic_read_u64(&XLogCtl->logFlushResult);
2592 : pg_read_barrier();
2593 : Write = pg_atomic_read_u64(&XLogCtl->logWriteResult);
2594 : pg_read_barrier();
2595 : Insert = pg_atomic_read_u64(&XLogCtl->logInsertResult);
2596 :
2597 : /* WAL written to disk is always ahead of WAL flushed */
2598 : Assert(Write >= Flush);
2599 :
2600 : /* WAL inserted to buffers is always ahead of WAL written */
2601 : Assert(Insert >= Write);
2602 : }
2603 : #endif
2604 4220304 : }
2605 :
2606 : /*
2607 : * Record the LSN for an asynchronous transaction commit/abort
2608 : * and nudge the WALWriter if there is work for it to do.
2609 : * (This should not be called for synchronous commits.)
2610 : */
2611 : void
2612 94892 : XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
2613 : {
2614 94892 : XLogRecPtr WriteRqstPtr = asyncXactLSN;
2615 : bool sleeping;
2616 94892 : bool wakeup = false;
2617 : XLogRecPtr prevAsyncXactLSN;
2618 :
2619 94892 : SpinLockAcquire(&XLogCtl->info_lck);
2620 94892 : sleeping = XLogCtl->WalWriterSleeping;
2621 94892 : prevAsyncXactLSN = XLogCtl->asyncXactLSN;
2622 94892 : if (XLogCtl->asyncXactLSN < asyncXactLSN)
2623 93988 : XLogCtl->asyncXactLSN = asyncXactLSN;
2624 94892 : SpinLockRelease(&XLogCtl->info_lck);
2625 :
2626 : /*
2627 : * If somebody else already called this function with a more aggressive
2628 : * LSN, they will have done what we needed (and perhaps more).
2629 : */
2630 94892 : if (asyncXactLSN <= prevAsyncXactLSN)
2631 904 : return;
2632 :
2633 : /*
2634 : * If the WALWriter is sleeping, kick it to make it come out of low-power
2635 : * mode, so that this async commit will reach disk within the expected
2636 : * amount of time. Otherwise, determine whether it has enough WAL
2637 : * available to flush, the same way that XLogBackgroundFlush() does.
2638 : */
2639 93988 : if (sleeping)
2640 30 : wakeup = true;
2641 : else
2642 : {
2643 : int flushblocks;
2644 :
2645 93958 : RefreshXLogWriteResult(LogwrtResult);
2646 :
2647 93958 : flushblocks =
2648 93958 : WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2649 :
2650 93958 : if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2651 7346 : wakeup = true;
2652 : }
2653 :
2654 93988 : if (wakeup)
2655 : {
2656 7376 : volatile PROC_HDR *procglobal = ProcGlobal;
2657 7376 : ProcNumber walwriterProc = procglobal->walwriterProc;
2658 :
2659 7376 : if (walwriterProc != INVALID_PROC_NUMBER)
2660 320 : SetLatch(&GetPGProcByNumber(walwriterProc)->procLatch);
2661 : }
2662 : }
2663 :
2664 : /*
2665 : * Record the LSN up to which we can remove WAL because it's not required by
2666 : * any replication slot.
2667 : */
2668 : void
2669 44536 : XLogSetReplicationSlotMinimumLSN(XLogRecPtr lsn)
2670 : {
2671 44536 : SpinLockAcquire(&XLogCtl->info_lck);
2672 44536 : XLogCtl->replicationSlotMinLSN = lsn;
2673 44536 : SpinLockRelease(&XLogCtl->info_lck);
2674 44536 : }
2675 :
2676 :
2677 : /*
2678 : * Return the oldest LSN we must retain to satisfy the needs of some
2679 : * replication slot.
2680 : */
2681 : static XLogRecPtr
2682 4424 : XLogGetReplicationSlotMinimumLSN(void)
2683 : {
2684 : XLogRecPtr retval;
2685 :
2686 4424 : SpinLockAcquire(&XLogCtl->info_lck);
2687 4424 : retval = XLogCtl->replicationSlotMinLSN;
2688 4424 : SpinLockRelease(&XLogCtl->info_lck);
2689 :
2690 4424 : return retval;
2691 : }
2692 :
2693 : /*
2694 : * Advance minRecoveryPoint in control file.
2695 : *
2696 : * If we crash during recovery, we must reach this point again before the
2697 : * database is consistent.
2698 : *
2699 : * If 'force' is true, 'lsn' argument is ignored. Otherwise, minRecoveryPoint
2700 : * is only updated if it's not already greater than or equal to 'lsn'.
2701 : */
2702 : static void
2703 215728 : UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
2704 : {
2705 : /* Quick check using our local copy of the variable */
2706 215728 : if (!updateMinRecoveryPoint || (!force && lsn <= LocalMinRecoveryPoint))
2707 202612 : return;
2708 :
2709 : /*
2710 : * An invalid minRecoveryPoint means that we need to recover all the WAL,
2711 : * i.e., we're doing crash recovery. We never modify the control file's
2712 : * value in that case, so we can short-circuit future checks here too. The
2713 : * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2714 : * updated until crash recovery finishes. We only do this for the startup
2715 : * process as it should not update its own reference of minRecoveryPoint
2716 : * until it has finished crash recovery to make sure that all WAL
2717 : * available is replayed in this case. This also saves from extra locks
2718 : * taken on the control file from the startup process.
2719 : */
2720 13116 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint) && InRecovery)
2721 : {
2722 62 : updateMinRecoveryPoint = false;
2723 62 : return;
2724 : }
2725 :
2726 13054 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2727 :
2728 : /* update local copy */
2729 13054 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
2730 13054 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
2731 :
2732 13054 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint))
2733 6 : updateMinRecoveryPoint = false;
2734 13048 : else if (force || LocalMinRecoveryPoint < lsn)
2735 : {
2736 : XLogRecPtr newMinRecoveryPoint;
2737 : TimeLineID newMinRecoveryPointTLI;
2738 :
2739 : /*
2740 : * To avoid having to update the control file too often, we update it
2741 : * all the way to the last record being replayed, even though 'lsn'
2742 : * would suffice for correctness. This also allows the 'force' case
2743 : * to not need a valid 'lsn' value.
2744 : *
2745 : * Another important reason for doing it this way is that the passed
2746 : * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2747 : * the caller got it from a corrupted heap page. Accepting such a
2748 : * value as the min recovery point would prevent us from coming up at
2749 : * all. Instead, we just log a warning and continue with recovery.
2750 : * (See also the comments about corrupt LSNs in XLogFlush.)
2751 : */
2752 10606 : newMinRecoveryPoint = GetCurrentReplayRecPtr(&newMinRecoveryPointTLI);
2753 10606 : if (!force && newMinRecoveryPoint < lsn)
2754 0 : elog(WARNING,
2755 : "xlog min recovery request %X/%08X is past current point %X/%08X",
2756 : LSN_FORMAT_ARGS(lsn), LSN_FORMAT_ARGS(newMinRecoveryPoint));
2757 :
2758 : /* update control file */
2759 10606 : if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2760 : {
2761 9912 : ControlFile->minRecoveryPoint = newMinRecoveryPoint;
2762 9912 : ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
2763 9912 : UpdateControlFile();
2764 9912 : LocalMinRecoveryPoint = newMinRecoveryPoint;
2765 9912 : LocalMinRecoveryPointTLI = newMinRecoveryPointTLI;
2766 :
2767 9912 : ereport(DEBUG2,
2768 : errmsg_internal("updated min recovery point to %X/%08X on timeline %u",
2769 : LSN_FORMAT_ARGS(newMinRecoveryPoint),
2770 : newMinRecoveryPointTLI));
2771 : }
2772 : }
2773 13054 : LWLockRelease(ControlFileLock);
2774 : }
2775 :
2776 : /*
2777 : * Ensure that all XLOG data through the given position is flushed to disk.
2778 : *
2779 : * NOTE: this differs from XLogWrite mainly in that the WALWriteLock is not
2780 : * already held, and we try to avoid acquiring it if possible.
2781 : */
2782 : void
2783 1388742 : XLogFlush(XLogRecPtr record)
2784 : {
2785 : XLogRecPtr WriteRqstPtr;
2786 : XLogwrtRqst WriteRqst;
2787 1388742 : TimeLineID insertTLI = XLogCtl->InsertTimeLineID;
2788 :
2789 : /*
2790 : * During REDO, we are reading not writing WAL. Therefore, instead of
2791 : * trying to flush the WAL, we should update minRecoveryPoint instead. We
2792 : * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2793 : * to act this way too, and because when it tries to write the
2794 : * end-of-recovery checkpoint, it should indeed flush.
2795 : */
2796 1388742 : if (!XLogInsertAllowed())
2797 : {
2798 214824 : UpdateMinRecoveryPoint(record, false);
2799 1109402 : return;
2800 : }
2801 :
2802 : /* Quick exit if already known flushed */
2803 1173918 : if (record <= LogwrtResult.Flush)
2804 894578 : return;
2805 :
2806 : #ifdef WAL_DEBUG
2807 : if (XLOG_DEBUG)
2808 : elog(LOG, "xlog flush request %X/%08X; write %X/%08X; flush %X/%08X",
2809 : LSN_FORMAT_ARGS(record),
2810 : LSN_FORMAT_ARGS(LogwrtResult.Write),
2811 : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2812 : #endif
2813 :
2814 279340 : START_CRIT_SECTION();
2815 :
2816 : /*
2817 : * Since fsync is usually a horribly expensive operation, we try to
2818 : * piggyback as much data as we can on each fsync: if we see any more data
2819 : * entered into the xlog buffer, we'll write and fsync that too, so that
2820 : * the final value of LogwrtResult.Flush is as large as possible. This
2821 : * gives us some chance of avoiding another fsync immediately after.
2822 : */
2823 :
2824 : /* initialize to given target; may increase below */
2825 279340 : WriteRqstPtr = record;
2826 :
2827 : /*
2828 : * Now wait until we get the write lock, or someone else does the flush
2829 : * for us.
2830 : */
2831 : for (;;)
2832 11990 : {
2833 : XLogRecPtr insertpos;
2834 :
2835 : /* done already? */
2836 291330 : RefreshXLogWriteResult(LogwrtResult);
2837 291330 : if (record <= LogwrtResult.Flush)
2838 21964 : break;
2839 :
2840 : /*
2841 : * Before actually performing the write, wait for all in-flight
2842 : * insertions to the pages we're about to write to finish.
2843 : */
2844 269366 : SpinLockAcquire(&XLogCtl->info_lck);
2845 269366 : if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2846 20868 : WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2847 269366 : SpinLockRelease(&XLogCtl->info_lck);
2848 269366 : insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2849 :
2850 : /*
2851 : * Try to get the write lock. If we can't get it immediately, wait
2852 : * until it's released, and recheck if we still need to do the flush
2853 : * or if the backend that held the lock did it for us already. This
2854 : * helps to maintain a good rate of group committing when the system
2855 : * is bottlenecked by the speed of fsyncing.
2856 : */
2857 269366 : if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2858 : {
2859 : /*
2860 : * The lock is now free, but we didn't acquire it yet. Before we
2861 : * do, loop back to check if someone else flushed the record for
2862 : * us already.
2863 : */
2864 11990 : continue;
2865 : }
2866 :
2867 : /* Got the lock; recheck whether request is satisfied */
2868 257376 : RefreshXLogWriteResult(LogwrtResult);
2869 257376 : if (record <= LogwrtResult.Flush)
2870 : {
2871 3568 : LWLockRelease(WALWriteLock);
2872 3568 : break;
2873 : }
2874 :
2875 : /*
2876 : * Sleep before flush! By adding a delay here, we may give further
2877 : * backends the opportunity to join the backlog of group commit
2878 : * followers; this can significantly improve transaction throughput,
2879 : * at the risk of increasing transaction latency.
2880 : *
2881 : * We do not sleep if enableFsync is not turned on, nor if there are
2882 : * fewer than CommitSiblings other backends with active transactions.
2883 : */
2884 253808 : if (CommitDelay > 0 && enableFsync &&
2885 0 : MinimumActiveBackends(CommitSiblings))
2886 : {
2887 0 : pg_usleep(CommitDelay);
2888 :
2889 : /*
2890 : * Re-check how far we can now flush the WAL. It's generally not
2891 : * safe to call WaitXLogInsertionsToFinish while holding
2892 : * WALWriteLock, because an in-progress insertion might need to
2893 : * also grab WALWriteLock to make progress. But we know that all
2894 : * the insertions up to insertpos have already finished, because
2895 : * that's what the earlier WaitXLogInsertionsToFinish() returned.
2896 : * We're only calling it again to allow insertpos to be moved
2897 : * further forward, not to actually wait for anyone.
2898 : */
2899 0 : insertpos = WaitXLogInsertionsToFinish(insertpos);
2900 : }
2901 :
2902 : /* try to write/flush later additions to XLOG as well */
2903 253808 : WriteRqst.Write = insertpos;
2904 253808 : WriteRqst.Flush = insertpos;
2905 :
2906 253808 : XLogWrite(WriteRqst, insertTLI, false);
2907 :
2908 253808 : LWLockRelease(WALWriteLock);
2909 : /* done */
2910 253808 : break;
2911 : }
2912 :
2913 279340 : END_CRIT_SECTION();
2914 :
2915 : /* wake up walsenders now that we've released heavily contended locks */
2916 279340 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
2917 :
2918 : /*
2919 : * If we still haven't flushed to the request point then we have a
2920 : * problem; most likely, the requested flush point is past end of XLOG.
2921 : * This has been seen to occur when a disk page has a corrupted LSN.
2922 : *
2923 : * Formerly we treated this as a PANIC condition, but that hurts the
2924 : * system's robustness rather than helping it: we do not want to take down
2925 : * the whole system due to corruption on one data page. In particular, if
2926 : * the bad page is encountered again during recovery then we would be
2927 : * unable to restart the database at all! (This scenario actually
2928 : * happened in the field several times with 7.1 releases.) As of 8.4, bad
2929 : * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2930 : * the only time we can reach here during recovery is while flushing the
2931 : * end-of-recovery checkpoint record, and we don't expect that to have a
2932 : * bad LSN.
2933 : *
2934 : * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2935 : * since xact.c calls this routine inside a critical section. However,
2936 : * calls from bufmgr.c are not within critical sections and so we will not
2937 : * force a restart for a bad LSN on a data page.
2938 : */
2939 279340 : if (LogwrtResult.Flush < record)
2940 0 : elog(ERROR,
2941 : "xlog flush request %X/%08X is not satisfied --- flushed only to %X/%08X",
2942 : LSN_FORMAT_ARGS(record),
2943 : LSN_FORMAT_ARGS(LogwrtResult.Flush));
2944 :
2945 : /*
2946 : * Cross-check XLogNeedsFlush(). Some of the checks of XLogFlush() and
2947 : * XLogNeedsFlush() are duplicated, and this assertion ensures that these
2948 : * remain consistent.
2949 : */
2950 : Assert(!XLogNeedsFlush(record));
2951 : }
2952 :
2953 : /*
2954 : * Write & flush xlog, but without specifying exactly where to.
2955 : *
2956 : * We normally write only completed blocks; but if there is nothing to do on
2957 : * that basis, we check for unwritten async commits in the current incomplete
2958 : * block, and write through the latest one of those. Thus, if async commits
2959 : * are not being used, we will write complete blocks only.
2960 : *
2961 : * If, based on the above, there's anything to write we do so immediately. But
2962 : * to avoid calling fsync, fdatasync et. al. at a rate that'd impact
2963 : * concurrent IO, we only flush WAL every wal_writer_delay ms, or if there's
2964 : * more than wal_writer_flush_after unflushed blocks.
2965 : *
2966 : * We can guarantee that async commits reach disk after at most three
2967 : * wal_writer_delay cycles. (When flushing complete blocks, we allow XLogWrite
2968 : * to write "flexibly", meaning it can stop at the end of the buffer ring;
2969 : * this makes a difference only with very high load or long wal_writer_delay,
2970 : * but imposes one extra cycle for the worst case for async commits.)
2971 : *
2972 : * This routine is invoked periodically by the background walwriter process.
2973 : *
2974 : * Returns true if there was any work to do, even if we skipped flushing due
2975 : * to wal_writer_delay/wal_writer_flush_after.
2976 : */
2977 : bool
2978 32642 : XLogBackgroundFlush(void)
2979 : {
2980 : XLogwrtRqst WriteRqst;
2981 32642 : bool flexible = true;
2982 : static TimestampTz lastflush;
2983 : TimestampTz now;
2984 : int flushblocks;
2985 : TimeLineID insertTLI;
2986 :
2987 : /* XLOG doesn't need flushing during recovery */
2988 32642 : if (RecoveryInProgress())
2989 12 : return false;
2990 :
2991 : /*
2992 : * Since we're not in recovery, InsertTimeLineID is set and can't change,
2993 : * so we can read it without a lock.
2994 : */
2995 32630 : insertTLI = XLogCtl->InsertTimeLineID;
2996 :
2997 : /* read updated LogwrtRqst */
2998 32630 : SpinLockAcquire(&XLogCtl->info_lck);
2999 32630 : WriteRqst = XLogCtl->LogwrtRqst;
3000 32630 : SpinLockRelease(&XLogCtl->info_lck);
3001 :
3002 : /* back off to last completed page boundary */
3003 32630 : WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3004 :
3005 : /* if we have already flushed that far, consider async commit records */
3006 32630 : RefreshXLogWriteResult(LogwrtResult);
3007 32630 : if (WriteRqst.Write <= LogwrtResult.Flush)
3008 : {
3009 24958 : SpinLockAcquire(&XLogCtl->info_lck);
3010 24958 : WriteRqst.Write = XLogCtl->asyncXactLSN;
3011 24958 : SpinLockRelease(&XLogCtl->info_lck);
3012 24958 : flexible = false; /* ensure it all gets written */
3013 : }
3014 :
3015 : /*
3016 : * If already known flushed, we're done. Just need to check if we are
3017 : * holding an open file handle to a logfile that's no longer in use,
3018 : * preventing the file from being deleted.
3019 : */
3020 32630 : if (WriteRqst.Write <= LogwrtResult.Flush)
3021 : {
3022 23660 : if (openLogFile >= 0)
3023 : {
3024 14414 : if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
3025 : wal_segment_size))
3026 : {
3027 240 : XLogFileClose();
3028 : }
3029 : }
3030 23660 : return false;
3031 : }
3032 :
3033 : /*
3034 : * Determine how far to flush WAL, based on the wal_writer_delay and
3035 : * wal_writer_flush_after GUCs.
3036 : *
3037 : * Note that XLogSetAsyncXactLSN() performs similar calculation based on
3038 : * wal_writer_flush_after, to decide when to wake us up. Make sure the
3039 : * logic is the same in both places if you change this.
3040 : */
3041 8970 : now = GetCurrentTimestamp();
3042 8970 : flushblocks =
3043 8970 : WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3044 :
3045 8970 : if (WalWriterFlushAfter == 0 || lastflush == 0)
3046 : {
3047 : /* first call, or block based limits disabled */
3048 506 : WriteRqst.Flush = WriteRqst.Write;
3049 506 : lastflush = now;
3050 : }
3051 8464 : else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3052 : {
3053 : /*
3054 : * Flush the writes at least every WalWriterDelay ms. This is
3055 : * important to bound the amount of time it takes for an asynchronous
3056 : * commit to hit disk.
3057 : */
3058 8246 : WriteRqst.Flush = WriteRqst.Write;
3059 8246 : lastflush = now;
3060 : }
3061 218 : else if (flushblocks >= WalWriterFlushAfter)
3062 : {
3063 : /* exceeded wal_writer_flush_after blocks, flush */
3064 198 : WriteRqst.Flush = WriteRqst.Write;
3065 198 : lastflush = now;
3066 : }
3067 : else
3068 : {
3069 : /* no flushing, this time round */
3070 20 : WriteRqst.Flush = 0;
3071 : }
3072 :
3073 : #ifdef WAL_DEBUG
3074 : if (XLOG_DEBUG)
3075 : elog(LOG, "xlog bg flush request write %X/%08X; flush: %X/%08X, current is write %X/%08X; flush %X/%08X",
3076 : LSN_FORMAT_ARGS(WriteRqst.Write),
3077 : LSN_FORMAT_ARGS(WriteRqst.Flush),
3078 : LSN_FORMAT_ARGS(LogwrtResult.Write),
3079 : LSN_FORMAT_ARGS(LogwrtResult.Flush));
3080 : #endif
3081 :
3082 8970 : START_CRIT_SECTION();
3083 :
3084 : /* now wait for any in-progress insertions to finish and get write lock */
3085 8970 : WaitXLogInsertionsToFinish(WriteRqst.Write);
3086 8970 : LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3087 8970 : RefreshXLogWriteResult(LogwrtResult);
3088 8970 : if (WriteRqst.Write > LogwrtResult.Write ||
3089 526 : WriteRqst.Flush > LogwrtResult.Flush)
3090 : {
3091 8590 : XLogWrite(WriteRqst, insertTLI, flexible);
3092 : }
3093 8970 : LWLockRelease(WALWriteLock);
3094 :
3095 8970 : END_CRIT_SECTION();
3096 :
3097 : /* wake up walsenders now that we've released heavily contended locks */
3098 8970 : WalSndWakeupProcessRequests(true, !RecoveryInProgress());
3099 :
3100 : /*
3101 : * Great, done. To take some work off the critical path, try to initialize
3102 : * as many of the no-longer-needed WAL buffers for future use as we can.
3103 : */
3104 8970 : AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3105 :
3106 : /*
3107 : * If we determined that we need to write data, but somebody else
3108 : * wrote/flushed already, it should be considered as being active, to
3109 : * avoid hibernating too early.
3110 : */
3111 8970 : return true;
3112 : }
3113 :
3114 : /*
3115 : * Test whether XLOG data has been flushed up to (at least) the given
3116 : * position, or whether the minimum recovery point has been updated past
3117 : * the given position.
3118 : *
3119 : * Returns true if a flush is still needed, or if the minimum recovery point
3120 : * must be updated.
3121 : *
3122 : * It is possible that someone else is already in the process of flushing
3123 : * that far, or has updated the minimum recovery point up to the given
3124 : * position.
3125 : */
3126 : bool
3127 17166316 : XLogNeedsFlush(XLogRecPtr record)
3128 : {
3129 : /*
3130 : * During recovery, we don't flush WAL but update minRecoveryPoint
3131 : * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3132 : * would need to be updated.
3133 : *
3134 : * Using XLogInsertAllowed() rather than RecoveryInProgress() matters for
3135 : * the case of an end-of-recovery checkpoint, where WAL data is flushed.
3136 : * This check should be consistent with the one in XLogFlush().
3137 : */
3138 17166316 : if (!XLogInsertAllowed())
3139 : {
3140 : /* Quick exit if already known to be updated or cannot be updated */
3141 1017314 : if (!updateMinRecoveryPoint || record <= LocalMinRecoveryPoint)
3142 999564 : return false;
3143 :
3144 : /*
3145 : * An invalid minRecoveryPoint means that we need to recover all the
3146 : * WAL, i.e., we're doing crash recovery. We never modify the control
3147 : * file's value in that case, so we can short-circuit future checks
3148 : * here too. This triggers a quick exit path for the startup process,
3149 : * which cannot update its local copy of minRecoveryPoint as long as
3150 : * it has not replayed all WAL available when doing crash recovery.
3151 : */
3152 17750 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint) && InRecovery)
3153 : {
3154 0 : updateMinRecoveryPoint = false;
3155 0 : return false;
3156 : }
3157 :
3158 : /*
3159 : * Update local copy of minRecoveryPoint. But if the lock is busy,
3160 : * just return a conservative guess.
3161 : */
3162 17750 : if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3163 0 : return true;
3164 17750 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
3165 17750 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
3166 17750 : LWLockRelease(ControlFileLock);
3167 :
3168 : /*
3169 : * Check minRecoveryPoint for any other process than the startup
3170 : * process doing crash recovery, which should not update the control
3171 : * file value if crash recovery is still running.
3172 : */
3173 17750 : if (!XLogRecPtrIsValid(LocalMinRecoveryPoint))
3174 0 : updateMinRecoveryPoint = false;
3175 :
3176 : /* check again */
3177 17750 : if (record <= LocalMinRecoveryPoint || !updateMinRecoveryPoint)
3178 166 : return false;
3179 : else
3180 17584 : return true;
3181 : }
3182 :
3183 : /* Quick exit if already known flushed */
3184 16149002 : if (record <= LogwrtResult.Flush)
3185 15693882 : return false;
3186 :
3187 : /* read LogwrtResult and update local state */
3188 455120 : RefreshXLogWriteResult(LogwrtResult);
3189 :
3190 : /* check again */
3191 455120 : if (record <= LogwrtResult.Flush)
3192 6474 : return false;
3193 :
3194 448646 : return true;
3195 : }
3196 :
3197 : /*
3198 : * Try to make a given XLOG file segment exist.
3199 : *
3200 : * logsegno: identify segment.
3201 : *
3202 : * *added: on return, true if this call raised the number of extant segments.
3203 : *
3204 : * path: on return, this char[MAXPGPATH] has the path to the logsegno file.
3205 : *
3206 : * Returns -1 or FD of opened file. A -1 here is not an error; a caller
3207 : * wanting an open segment should attempt to open "path", which usually will
3208 : * succeed. (This is weird, but it's efficient for the callers.)
3209 : */
3210 : static int
3211 28824 : XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli,
3212 : bool *added, char *path)
3213 : {
3214 : char tmppath[MAXPGPATH];
3215 : XLogSegNo installed_segno;
3216 : XLogSegNo max_segno;
3217 : int fd;
3218 : int save_errno;
3219 28824 : int open_flags = O_RDWR | O_CREAT | O_EXCL | PG_BINARY;
3220 : instr_time io_start;
3221 :
3222 : Assert(logtli != 0);
3223 :
3224 28824 : XLogFilePath(path, logtli, logsegno, wal_segment_size);
3225 :
3226 : /*
3227 : * Try to use existent file (checkpoint maker may have created it already)
3228 : */
3229 28824 : *added = false;
3230 28824 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3231 28824 : get_sync_bit(wal_sync_method));
3232 28824 : if (fd < 0)
3233 : {
3234 2986 : if (errno != ENOENT)
3235 0 : ereport(ERROR,
3236 : (errcode_for_file_access(),
3237 : errmsg("could not open file \"%s\": %m", path)));
3238 : }
3239 : else
3240 25838 : return fd;
3241 :
3242 : /*
3243 : * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3244 : * another process is doing the same thing. If so, we will end up
3245 : * pre-creating an extra log segment. That seems OK, and better than
3246 : * holding the lock throughout this lengthy process.
3247 : */
3248 2986 : elog(DEBUG2, "creating and filling new WAL file");
3249 :
3250 2986 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3251 :
3252 2986 : unlink(tmppath);
3253 :
3254 2986 : if (io_direct_flags & IO_DIRECT_WAL_INIT)
3255 0 : open_flags |= PG_O_DIRECT;
3256 :
3257 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3258 2986 : fd = BasicOpenFile(tmppath, open_flags);
3259 2986 : if (fd < 0)
3260 0 : ereport(ERROR,
3261 : (errcode_for_file_access(),
3262 : errmsg("could not create file \"%s\": %m", tmppath)));
3263 :
3264 : /* Measure I/O timing when initializing segment */
3265 2986 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3266 :
3267 2986 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3268 2986 : save_errno = 0;
3269 2986 : if (wal_init_zero)
3270 : {
3271 : ssize_t rc;
3272 :
3273 : /*
3274 : * Zero-fill the file. With this setting, we do this the hard way to
3275 : * ensure that all the file space has really been allocated. On
3276 : * platforms that allow "holes" in files, just seeking to the end
3277 : * doesn't allocate intermediate space. This way, we know that we
3278 : * have all the space and (after the fsync below) that all the
3279 : * indirect blocks are down on disk. Therefore, fdatasync(2) or
3280 : * O_DSYNC will be sufficient to sync future writes to the log file.
3281 : */
3282 2986 : rc = pg_pwrite_zeros(fd, wal_segment_size, 0);
3283 :
3284 2986 : if (rc < 0)
3285 0 : save_errno = errno;
3286 : }
3287 : else
3288 : {
3289 : /*
3290 : * Otherwise, seeking to the end and writing a solitary byte is
3291 : * enough.
3292 : */
3293 0 : errno = 0;
3294 0 : if (pg_pwrite(fd, "\0", 1, wal_segment_size - 1) != 1)
3295 : {
3296 : /* if write didn't set errno, assume no disk space */
3297 0 : save_errno = errno ? errno : ENOSPC;
3298 : }
3299 : }
3300 2986 : pgstat_report_wait_end();
3301 :
3302 : /*
3303 : * A full segment worth of data is written when using wal_init_zero. One
3304 : * byte is written when not using it.
3305 : */
3306 2986 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT, IOOP_WRITE,
3307 : io_start, 1,
3308 2986 : wal_init_zero ? wal_segment_size : 1);
3309 :
3310 2986 : if (save_errno)
3311 : {
3312 : /*
3313 : * If we fail to make the file, delete it to release disk space
3314 : */
3315 0 : unlink(tmppath);
3316 :
3317 0 : close(fd);
3318 :
3319 0 : errno = save_errno;
3320 :
3321 0 : ereport(ERROR,
3322 : (errcode_for_file_access(),
3323 : errmsg("could not write to file \"%s\": %m", tmppath)));
3324 : }
3325 :
3326 : /* Measure I/O timing when flushing segment */
3327 2986 : io_start = pgstat_prepare_io_time(track_wal_io_timing);
3328 :
3329 2986 : pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
3330 2986 : if (pg_fsync(fd) != 0)
3331 : {
3332 0 : save_errno = errno;
3333 0 : close(fd);
3334 0 : errno = save_errno;
3335 0 : ereport(ERROR,
3336 : (errcode_for_file_access(),
3337 : errmsg("could not fsync file \"%s\": %m", tmppath)));
3338 : }
3339 2986 : pgstat_report_wait_end();
3340 :
3341 2986 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_INIT,
3342 : IOOP_FSYNC, io_start, 1, 0);
3343 :
3344 2986 : if (close(fd) != 0)
3345 0 : ereport(ERROR,
3346 : (errcode_for_file_access(),
3347 : errmsg("could not close file \"%s\": %m", tmppath)));
3348 :
3349 : /*
3350 : * Now move the segment into place with its final name. Cope with
3351 : * possibility that someone else has created the file while we were
3352 : * filling ours: if so, use ours to pre-create a future log segment.
3353 : */
3354 2986 : installed_segno = logsegno;
3355 :
3356 : /*
3357 : * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3358 : * that was a constant, but that was always a bit dubious: normally, at a
3359 : * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3360 : * here, it was the offset from the insert location. We can't do the
3361 : * normal XLOGfileslop calculation here because we don't have access to
3362 : * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3363 : * CheckPointSegments.
3364 : */
3365 2986 : max_segno = logsegno + CheckPointSegments;
3366 2986 : if (InstallXLogFileSegment(&installed_segno, tmppath, true, max_segno,
3367 : logtli))
3368 : {
3369 2986 : *added = true;
3370 2986 : elog(DEBUG2, "done creating and filling new WAL file");
3371 : }
3372 : else
3373 : {
3374 : /*
3375 : * No need for any more future segments, or InstallXLogFileSegment()
3376 : * failed to rename the file into place. If the rename failed, a
3377 : * caller opening the file may fail.
3378 : */
3379 0 : unlink(tmppath);
3380 0 : elog(DEBUG2, "abandoned new WAL file");
3381 : }
3382 :
3383 2986 : return -1;
3384 : }
3385 :
3386 : /*
3387 : * Create a new XLOG file segment, or open a pre-existing one.
3388 : *
3389 : * logsegno: identify segment to be created/opened.
3390 : *
3391 : * Returns FD of opened file.
3392 : *
3393 : * Note: errors here are ERROR not PANIC because we might or might not be
3394 : * inside a critical section (eg, during checkpoint there is no reason to
3395 : * take down the system on failure). They will promote to PANIC if we are
3396 : * in a critical section.
3397 : */
3398 : int
3399 28306 : XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
3400 : {
3401 : bool ignore_added;
3402 : char path[MAXPGPATH];
3403 : int fd;
3404 :
3405 : Assert(logtli != 0);
3406 :
3407 28306 : fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
3408 28306 : if (fd >= 0)
3409 25608 : return fd;
3410 :
3411 : /* Now open original target segment (might not be file I just made) */
3412 2698 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3413 2698 : get_sync_bit(wal_sync_method));
3414 2698 : if (fd < 0)
3415 0 : ereport(ERROR,
3416 : (errcode_for_file_access(),
3417 : errmsg("could not open file \"%s\": %m", path)));
3418 2698 : return fd;
3419 : }
3420 :
3421 : /*
3422 : * Create a new XLOG file segment by copying a pre-existing one.
3423 : *
3424 : * destsegno: identify segment to be created.
3425 : *
3426 : * srcTLI, srcsegno: identify segment to be copied (could be from
3427 : * a different timeline)
3428 : *
3429 : * upto: how much of the source file to copy (the rest is filled with
3430 : * zeros)
3431 : *
3432 : * Currently this is only used during recovery, and so there are no locking
3433 : * considerations. But we should be just as tense as XLogFileInit to avoid
3434 : * emplacing a bogus file.
3435 : */
3436 : static void
3437 84 : XLogFileCopy(TimeLineID destTLI, XLogSegNo destsegno,
3438 : 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 84 : XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
3452 84 : srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
3453 84 : 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 84 : snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3462 :
3463 84 : unlink(tmppath);
3464 :
3465 : /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3466 84 : fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3467 84 : 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 172116 : for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3476 : {
3477 : int nread;
3478 :
3479 172032 : nread = upto - nbytes;
3480 :
3481 : /*
3482 : * The part that is not read from the source file is filled with
3483 : * zeros.
3484 : */
3485 172032 : if (nread < sizeof(buffer))
3486 84 : memset(buffer.data, 0, sizeof(buffer));
3487 :
3488 172032 : if (nread > 0)
3489 : {
3490 : int r;
3491 :
3492 5392 : if (nread > sizeof(buffer))
3493 5308 : nread = sizeof(buffer);
3494 5392 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
3495 5392 : r = read(srcfd, buffer.data, nread);
3496 5392 : 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 5392 : pgstat_report_wait_end();
3510 : }
3511 172032 : errno = 0;
3512 172032 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
3513 172032 : 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 172032 : pgstat_report_wait_end();
3529 : }
3530 :
3531 84 : pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
3532 84 : 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 84 : pgstat_report_wait_end();
3537 :
3538 84 : if (CloseTransientFile(fd) != 0)
3539 0 : ereport(ERROR,
3540 : (errcode_for_file_access(),
3541 : errmsg("could not close file \"%s\": %m", tmppath)));
3542 :
3543 84 : 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 84 : if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, destTLI))
3552 0 : elog(ERROR, "InstallXLogFileSegment should not have failed");
3553 84 : }
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 : * tli: The timeline on which the new segment should be installed.
3576 : *
3577 : * Returns true if the file was installed successfully. false indicates that
3578 : * max_segno limit was exceeded, the startup process has disabled this
3579 : * function for now, or an error occurred while renaming the file into place.
3580 : */
3581 : static bool
3582 5788 : InstallXLogFileSegment(XLogSegNo *segno, char *tmppath,
3583 : bool find_free, XLogSegNo max_segno, TimeLineID tli)
3584 : {
3585 : char path[MAXPGPATH];
3586 : struct stat stat_buf;
3587 :
3588 : Assert(tli != 0);
3589 :
3590 5788 : XLogFilePath(path, tli, *segno, wal_segment_size);
3591 :
3592 5788 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3593 5788 : if (!XLogCtl->InstallXLogFileSegmentActive)
3594 : {
3595 0 : LWLockRelease(ControlFileLock);
3596 0 : return false;
3597 : }
3598 :
3599 5788 : if (!find_free)
3600 : {
3601 : /* Force installation: get rid of any pre-existing segment file */
3602 84 : durable_unlink(path, DEBUG1);
3603 : }
3604 : else
3605 : {
3606 : /* Find a free slot to put it in */
3607 8260 : while (stat(path, &stat_buf) == 0)
3608 : {
3609 2630 : if ((*segno) >= max_segno)
3610 : {
3611 : /* Failed to find a free slot within specified range */
3612 74 : LWLockRelease(ControlFileLock);
3613 74 : return false;
3614 : }
3615 2556 : (*segno)++;
3616 2556 : XLogFilePath(path, tli, *segno, wal_segment_size);
3617 : }
3618 : }
3619 :
3620 : Assert(access(path, F_OK) != 0 && errno == ENOENT);
3621 5714 : if (durable_rename(tmppath, path, LOG) != 0)
3622 : {
3623 0 : LWLockRelease(ControlFileLock);
3624 : /* durable_rename already emitted log message */
3625 0 : return false;
3626 : }
3627 :
3628 5714 : LWLockRelease(ControlFileLock);
3629 :
3630 5714 : return true;
3631 : }
3632 :
3633 : /*
3634 : * Open a pre-existing logfile segment for writing.
3635 : */
3636 : int
3637 152 : XLogFileOpen(XLogSegNo segno, TimeLineID tli)
3638 : {
3639 : char path[MAXPGPATH];
3640 : int fd;
3641 :
3642 152 : XLogFilePath(path, tli, segno, wal_segment_size);
3643 :
3644 152 : fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3645 152 : get_sync_bit(wal_sync_method));
3646 152 : if (fd < 0)
3647 0 : ereport(PANIC,
3648 : (errcode_for_file_access(),
3649 : errmsg("could not open file \"%s\": %m", path)));
3650 :
3651 152 : return fd;
3652 : }
3653 :
3654 : /*
3655 : * Close the current logfile segment for writing.
3656 : */
3657 : static void
3658 12262 : XLogFileClose(void)
3659 : {
3660 : Assert(openLogFile >= 0);
3661 :
3662 : /*
3663 : * WAL segment files will not be re-read in normal operation, so we advise
3664 : * the OS to release any cached pages. But do not do so if WAL archiving
3665 : * or streaming is active, because archiver and walsender process could
3666 : * use the cache to read the WAL segment.
3667 : */
3668 : #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
3669 12262 : if (!XLogIsNeeded() && (io_direct_flags & IO_DIRECT_WAL) == 0)
3670 2752 : (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3671 : #endif
3672 :
3673 12262 : if (close(openLogFile) != 0)
3674 : {
3675 : char xlogfname[MAXFNAMELEN];
3676 0 : int save_errno = errno;
3677 :
3678 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo, wal_segment_size);
3679 0 : errno = save_errno;
3680 0 : ereport(PANIC,
3681 : (errcode_for_file_access(),
3682 : errmsg("could not close file \"%s\": %m", xlogfname)));
3683 : }
3684 :
3685 12262 : openLogFile = -1;
3686 12262 : ReleaseExternalFD();
3687 12262 : }
3688 :
3689 : /*
3690 : * Preallocate log files beyond the specified log endpoint.
3691 : *
3692 : * XXX this is currently extremely conservative, since it forces only one
3693 : * future log segment to exist, and even that only if we are 75% done with
3694 : * the current one. This is only appropriate for very low-WAL-volume systems.
3695 : * High-volume systems will be OK once they've built up a sufficient set of
3696 : * recycled log segments, but the startup transient is likely to include
3697 : * a lot of segment creations by foreground processes, which is not so good.
3698 : *
3699 : * XLogFileInitInternal() can ereport(ERROR). All known causes indicate big
3700 : * trouble; for example, a full filesystem is one cause. The checkpoint WAL
3701 : * and/or ControlFile updates already completed. If a RequestCheckpoint()
3702 : * initiated the present checkpoint and an ERROR ends this function, the
3703 : * command that called RequestCheckpoint() fails. That's not ideal, but it's
3704 : * not worth contorting more functions to use caller-specified elevel values.
3705 : * (With or without RequestCheckpoint(), an ERROR forestalls some inessential
3706 : * reporting and resource reclamation.)
3707 : */
3708 : static void
3709 3992 : PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
3710 : {
3711 : XLogSegNo _logSegNo;
3712 : int lf;
3713 : bool added;
3714 : char path[MAXPGPATH];
3715 : uint64 offset;
3716 :
3717 3992 : if (!XLogCtl->InstallXLogFileSegmentActive)
3718 20 : return; /* unlocked check says no */
3719 :
3720 3972 : XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3721 3972 : offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3722 3972 : if (offset >= (uint32) (0.75 * wal_segment_size))
3723 : {
3724 518 : _logSegNo++;
3725 518 : lf = XLogFileInitInternal(_logSegNo, tli, &added, path);
3726 518 : if (lf >= 0)
3727 230 : close(lf);
3728 518 : if (added)
3729 288 : CheckpointStats.ckpt_segs_added++;
3730 : }
3731 : }
3732 :
3733 : /*
3734 : * Throws an error if the given log segment has already been removed or
3735 : * recycled. The caller should only pass a segment that it knows to have
3736 : * existed while the server has been running, as this function always
3737 : * succeeds if no WAL segments have been removed since startup.
3738 : * 'tli' is only used in the error message.
3739 : *
3740 : * Note: this function guarantees to keep errno unchanged on return.
3741 : * This supports callers that use this to possibly deliver a better
3742 : * error message about a missing file, while still being able to throw
3743 : * a normal file-access error afterwards, if this does return.
3744 : */
3745 : void
3746 238120 : CheckXLogRemoved(XLogSegNo segno, TimeLineID tli)
3747 : {
3748 238120 : int save_errno = errno;
3749 : XLogSegNo lastRemovedSegNo;
3750 :
3751 238120 : SpinLockAcquire(&XLogCtl->info_lck);
3752 238120 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3753 238120 : SpinLockRelease(&XLogCtl->info_lck);
3754 :
3755 238120 : if (segno <= lastRemovedSegNo)
3756 : {
3757 : char filename[MAXFNAMELEN];
3758 :
3759 0 : XLogFileName(filename, tli, segno, wal_segment_size);
3760 0 : errno = save_errno;
3761 0 : ereport(ERROR,
3762 : (errcode_for_file_access(),
3763 : errmsg("requested WAL segment %s has already been removed",
3764 : filename)));
3765 : }
3766 238120 : errno = save_errno;
3767 238120 : }
3768 :
3769 : /*
3770 : * Return the last WAL segment removed, or 0 if no segment has been removed
3771 : * since startup.
3772 : *
3773 : * NB: the result can be out of date arbitrarily fast, the caller has to deal
3774 : * with that.
3775 : */
3776 : XLogSegNo
3777 2172 : XLogGetLastRemovedSegno(void)
3778 : {
3779 : XLogSegNo lastRemovedSegNo;
3780 :
3781 2172 : SpinLockAcquire(&XLogCtl->info_lck);
3782 2172 : lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3783 2172 : SpinLockRelease(&XLogCtl->info_lck);
3784 :
3785 2172 : return lastRemovedSegNo;
3786 : }
3787 :
3788 : /*
3789 : * Return the oldest WAL segment on the given TLI that still exists in
3790 : * XLOGDIR, or 0 if none.
3791 : */
3792 : XLogSegNo
3793 10 : XLogGetOldestSegno(TimeLineID tli)
3794 : {
3795 : DIR *xldir;
3796 : struct dirent *xlde;
3797 10 : XLogSegNo oldest_segno = 0;
3798 :
3799 10 : xldir = AllocateDir(XLOGDIR);
3800 68 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3801 : {
3802 : TimeLineID file_tli;
3803 : XLogSegNo file_segno;
3804 :
3805 : /* Ignore files that are not XLOG segments. */
3806 58 : if (!IsXLogFileName(xlde->d_name))
3807 40 : continue;
3808 :
3809 : /* Parse filename to get TLI and segno. */
3810 18 : XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3811 : wal_segment_size);
3812 :
3813 : /* Ignore anything that's not from the TLI of interest. */
3814 18 : if (tli != file_tli)
3815 0 : continue;
3816 :
3817 : /* If it's the oldest so far, update oldest_segno. */
3818 18 : if (oldest_segno == 0 || file_segno < oldest_segno)
3819 16 : oldest_segno = file_segno;
3820 : }
3821 :
3822 10 : FreeDir(xldir);
3823 10 : return oldest_segno;
3824 : }
3825 :
3826 : /*
3827 : * Update the last removed segno pointer in shared memory, to reflect that the
3828 : * given XLOG file has been removed.
3829 : */
3830 : static void
3831 5072 : UpdateLastRemovedPtr(char *filename)
3832 : {
3833 : uint32 tli;
3834 : XLogSegNo segno;
3835 :
3836 5072 : XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3837 :
3838 5072 : SpinLockAcquire(&XLogCtl->info_lck);
3839 5072 : if (segno > XLogCtl->lastRemovedSegNo)
3840 2126 : XLogCtl->lastRemovedSegNo = segno;
3841 5072 : SpinLockRelease(&XLogCtl->info_lck);
3842 5072 : }
3843 :
3844 : /*
3845 : * Remove all temporary log files in pg_wal
3846 : *
3847 : * This is called at the beginning of recovery after a previous crash,
3848 : * at a point where no other processes write fresh WAL data.
3849 : */
3850 : static void
3851 354 : RemoveTempXlogFiles(void)
3852 : {
3853 : DIR *xldir;
3854 : struct dirent *xlde;
3855 :
3856 354 : elog(DEBUG2, "removing all temporary WAL segments");
3857 :
3858 354 : xldir = AllocateDir(XLOGDIR);
3859 2406 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3860 : {
3861 : char path[MAXPGPATH];
3862 :
3863 2052 : if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3864 2052 : continue;
3865 :
3866 0 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3867 0 : unlink(path);
3868 0 : elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3869 : }
3870 354 : FreeDir(xldir);
3871 354 : }
3872 :
3873 : /*
3874 : * Recycle or remove all log files older or equal to passed segno.
3875 : *
3876 : * endptr is current (or recent) end of xlog, and lastredoptr is the
3877 : * redo pointer of the last checkpoint. These are used to determine
3878 : * whether we want to recycle rather than delete no-longer-wanted log files.
3879 : *
3880 : * insertTLI is the current timeline for XLOG insertion. Any recycled
3881 : * segments should be reused for this timeline.
3882 : */
3883 : static void
3884 3458 : RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr,
3885 : TimeLineID insertTLI)
3886 : {
3887 : DIR *xldir;
3888 : struct dirent *xlde;
3889 : char lastoff[MAXFNAMELEN];
3890 : XLogSegNo endlogSegNo;
3891 : XLogSegNo recycleSegNo;
3892 :
3893 : /* Initialize info about where to try to recycle to */
3894 3458 : XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
3895 3458 : recycleSegNo = XLOGfileslop(lastredoptr);
3896 :
3897 : /*
3898 : * Construct a filename of the last segment to be kept. The timeline ID
3899 : * doesn't matter, we ignore that in the comparison. (During recovery,
3900 : * InsertTimeLineID isn't set, so we can't use that.)
3901 : */
3902 3458 : XLogFileName(lastoff, 0, segno, wal_segment_size);
3903 :
3904 3458 : elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3905 : lastoff);
3906 :
3907 3458 : xldir = AllocateDir(XLOGDIR);
3908 :
3909 87966 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3910 : {
3911 : /* Ignore files that are not XLOG segments */
3912 84508 : if (!IsXLogFileName(xlde->d_name) &&
3913 14950 : !IsPartialXLogFileName(xlde->d_name))
3914 14942 : continue;
3915 :
3916 : /*
3917 : * We ignore the timeline part of the XLOG segment identifiers in
3918 : * deciding whether a segment is still needed. This ensures that we
3919 : * won't prematurely remove a segment from a parent timeline. We could
3920 : * probably be a little more proactive about removing segments of
3921 : * non-parent timelines, but that would be a whole lot more
3922 : * complicated.
3923 : *
3924 : * We use the alphanumeric sorting property of the filenames to decide
3925 : * which ones are earlier than the lastoff segment.
3926 : */
3927 69566 : if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3928 : {
3929 43198 : if (XLogArchiveCheckDone(xlde->d_name))
3930 : {
3931 : /* Update the last removed location in shared memory first */
3932 5072 : UpdateLastRemovedPtr(xlde->d_name);
3933 :
3934 5072 : RemoveXlogFile(xlde, recycleSegNo, &endlogSegNo, insertTLI);
3935 : }
3936 : }
3937 : }
3938 :
3939 3458 : FreeDir(xldir);
3940 3458 : }
3941 :
3942 : /*
3943 : * Recycle or remove WAL files that are not part of the given timeline's
3944 : * history.
3945 : *
3946 : * This is called during recovery, whenever we switch to follow a new
3947 : * timeline, and at the end of recovery when we create a new timeline. We
3948 : * wouldn't otherwise care about extra WAL files lying in pg_wal, but they
3949 : * might be leftover pre-allocated or recycled WAL segments on the old timeline
3950 : * that we haven't used yet, and contain garbage. If we just leave them in
3951 : * pg_wal, they will eventually be archived, and we can't let that happen.
3952 : * Files that belong to our timeline history are valid, because we have
3953 : * successfully replayed them, but from others we can't be sure.
3954 : *
3955 : * 'switchpoint' is the current point in WAL where we switch to new timeline,
3956 : * and 'newTLI' is the new timeline we switch to.
3957 : */
3958 : void
3959 124 : RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
3960 : {
3961 : DIR *xldir;
3962 : struct dirent *xlde;
3963 : char switchseg[MAXFNAMELEN];
3964 : XLogSegNo endLogSegNo;
3965 : XLogSegNo switchLogSegNo;
3966 : XLogSegNo recycleSegNo;
3967 :
3968 : /*
3969 : * Initialize info about where to begin the work. This will recycle,
3970 : * somewhat arbitrarily, 10 future segments.
3971 : */
3972 124 : XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3973 124 : XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3974 124 : recycleSegNo = endLogSegNo + 10;
3975 :
3976 : /*
3977 : * Construct a filename of the last segment to be kept.
3978 : */
3979 124 : XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3980 :
3981 124 : elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3982 : switchseg);
3983 :
3984 124 : xldir = AllocateDir(XLOGDIR);
3985 :
3986 1188 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3987 : {
3988 : /* Ignore files that are not XLOG segments */
3989 1064 : if (!IsXLogFileName(xlde->d_name))
3990 656 : continue;
3991 :
3992 : /*
3993 : * Remove files that are on a timeline older than the new one we're
3994 : * switching to, but with a segment number >= the first segment on the
3995 : * new timeline.
3996 : */
3997 408 : if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3998 268 : strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3999 : {
4000 : /*
4001 : * If the file has already been marked as .ready, however, don't
4002 : * remove it yet. It should be OK to remove it - files that are
4003 : * not part of our timeline history are not required for recovery
4004 : * - but seems safer to let them be archived and removed later.
4005 : */
4006 34 : if (!XLogArchiveIsReady(xlde->d_name))
4007 34 : RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
4008 : }
4009 : }
4010 :
4011 124 : FreeDir(xldir);
4012 124 : }
4013 :
4014 : /*
4015 : * Recycle or remove a log file that's no longer needed.
4016 : *
4017 : * segment_de is the dirent structure of the segment to recycle or remove.
4018 : * recycleSegNo is the segment number to recycle up to. endlogSegNo is
4019 : * the segment number of the current (or recent) end of WAL.
4020 : *
4021 : * endlogSegNo gets incremented if the segment is recycled so as it is not
4022 : * checked again with future callers of this function.
4023 : *
4024 : * insertTLI is the current timeline for XLOG insertion. Any recycled segments
4025 : * should be used for this timeline.
4026 : */
4027 : static void
4028 5106 : RemoveXlogFile(const struct dirent *segment_de,
4029 : XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo,
4030 : TimeLineID insertTLI)
4031 : {
4032 : char path[MAXPGPATH];
4033 : #ifdef WIN32
4034 : char newpath[MAXPGPATH];
4035 : #endif
4036 5106 : const char *segname = segment_de->d_name;
4037 :
4038 5106 : snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4039 :
4040 : /*
4041 : * Before deleting the file, see if it can be recycled as a future log
4042 : * segment. Only recycle normal files, because we don't want to recycle
4043 : * symbolic links pointing to a separate archive directory.
4044 : */
4045 5106 : if (wal_recycle &&
4046 5106 : *endlogSegNo <= recycleSegNo &&
4047 6088 : XLogCtl->InstallXLogFileSegmentActive && /* callee rechecks this */
4048 5436 : get_dirent_type(path, segment_de, false, DEBUG2) == PGFILETYPE_REG &&
4049 2718 : InstallXLogFileSegment(endlogSegNo, path,
4050 : true, recycleSegNo, insertTLI))
4051 : {
4052 2644 : ereport(DEBUG2,
4053 : (errmsg_internal("recycled write-ahead log file \"%s\"",
4054 : segname)));
4055 2644 : CheckpointStats.ckpt_segs_recycled++;
4056 : /* Needn't recheck that slot on future iterations */
4057 2644 : (*endlogSegNo)++;
4058 : }
4059 : else
4060 : {
4061 : /* No need for any more future segments, or recycling failed ... */
4062 : int rc;
4063 :
4064 2462 : ereport(DEBUG2,
4065 : (errmsg_internal("removing write-ahead log file \"%s\"",
4066 : segname)));
4067 :
4068 : #ifdef WIN32
4069 :
4070 : /*
4071 : * On Windows, if another process (e.g another backend) holds the file
4072 : * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4073 : * will still show up in directory listing until the last handle is
4074 : * closed. To avoid confusing the lingering deleted file for a live
4075 : * WAL file that needs to be archived, rename it before deleting it.
4076 : *
4077 : * If another process holds the file open without FILE_SHARE_DELETE
4078 : * flag, rename will fail. We'll try again at the next checkpoint.
4079 : */
4080 : snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4081 : if (rename(path, newpath) != 0)
4082 : {
4083 : ereport(LOG,
4084 : (errcode_for_file_access(),
4085 : errmsg("could not rename file \"%s\": %m",
4086 : path)));
4087 : return;
4088 : }
4089 : rc = durable_unlink(newpath, LOG);
4090 : #else
4091 2462 : rc = durable_unlink(path, LOG);
4092 : #endif
4093 2462 : if (rc != 0)
4094 : {
4095 : /* Message already logged by durable_unlink() */
4096 0 : return;
4097 : }
4098 2462 : CheckpointStats.ckpt_segs_removed++;
4099 : }
4100 :
4101 5106 : XLogArchiveCleanup(segname);
4102 : }
4103 :
4104 : /*
4105 : * Verify whether pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
4106 : * If the latter do not exist, recreate them.
4107 : *
4108 : * It is not the goal of this function to verify the contents of these
4109 : * directories, but to help in cases where someone has performed a cluster
4110 : * copy for PITR purposes but omitted pg_wal from the copy.
4111 : *
4112 : * We could also recreate pg_wal if it doesn't exist, but a deliberate
4113 : * policy decision was made not to. It is fairly common for pg_wal to be
4114 : * a symlink, and if that was the DBA's intent then automatically making a
4115 : * plain directory would result in degraded performance with no notice.
4116 : */
4117 : static void
4118 1914 : ValidateXLOGDirectoryStructure(void)
4119 : {
4120 : char path[MAXPGPATH];
4121 : struct stat stat_buf;
4122 :
4123 : /* Check for pg_wal; if it doesn't exist, error out */
4124 1914 : if (stat(XLOGDIR, &stat_buf) != 0 ||
4125 1914 : !S_ISDIR(stat_buf.st_mode))
4126 0 : ereport(FATAL,
4127 : (errcode_for_file_access(),
4128 : errmsg("required WAL directory \"%s\" does not exist",
4129 : XLOGDIR)));
4130 :
4131 : /* Check for archive_status */
4132 1914 : snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4133 1914 : if (stat(path, &stat_buf) == 0)
4134 : {
4135 : /* Check for weird cases where it exists but isn't a directory */
4136 1912 : if (!S_ISDIR(stat_buf.st_mode))
4137 0 : ereport(FATAL,
4138 : (errcode_for_file_access(),
4139 : errmsg("required WAL directory \"%s\" does not exist",
4140 : path)));
4141 : }
4142 : else
4143 : {
4144 2 : ereport(LOG,
4145 : (errmsg("creating missing WAL directory \"%s\"", path)));
4146 2 : if (MakePGDirectory(path) < 0)
4147 0 : ereport(FATAL,
4148 : (errcode_for_file_access(),
4149 : errmsg("could not create missing directory \"%s\": %m",
4150 : path)));
4151 : }
4152 :
4153 : /* Check for summaries */
4154 1914 : snprintf(path, MAXPGPATH, XLOGDIR "/summaries");
4155 1914 : if (stat(path, &stat_buf) == 0)
4156 : {
4157 : /* Check for weird cases where it exists but isn't a directory */
4158 1912 : if (!S_ISDIR(stat_buf.st_mode))
4159 0 : ereport(FATAL,
4160 : (errmsg("required WAL directory \"%s\" does not exist",
4161 : path)));
4162 : }
4163 : else
4164 : {
4165 2 : ereport(LOG,
4166 : (errmsg("creating missing WAL directory \"%s\"", path)));
4167 2 : if (MakePGDirectory(path) < 0)
4168 0 : ereport(FATAL,
4169 : (errmsg("could not create missing directory \"%s\": %m",
4170 : path)));
4171 : }
4172 1914 : }
4173 :
4174 : /*
4175 : * Remove previous backup history files. This also retries creation of
4176 : * .ready files for any backup history files for which XLogArchiveNotify
4177 : * failed earlier.
4178 : */
4179 : static void
4180 300 : CleanupBackupHistory(void)
4181 : {
4182 : DIR *xldir;
4183 : struct dirent *xlde;
4184 : char path[MAXPGPATH + sizeof(XLOGDIR)];
4185 :
4186 300 : xldir = AllocateDir(XLOGDIR);
4187 :
4188 3016 : while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4189 : {
4190 2416 : if (IsBackupHistoryFileName(xlde->d_name))
4191 : {
4192 316 : if (XLogArchiveCheckDone(xlde->d_name))
4193 : {
4194 250 : elog(DEBUG2, "removing WAL backup history file \"%s\"",
4195 : xlde->d_name);
4196 250 : snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
4197 250 : unlink(path);
4198 250 : XLogArchiveCleanup(xlde->d_name);
4199 : }
4200 : }
4201 : }
4202 :
4203 300 : FreeDir(xldir);
4204 300 : }
4205 :
4206 : /*
4207 : * I/O routines for pg_control
4208 : *
4209 : * *ControlFile is a buffer in shared memory that holds an image of the
4210 : * contents of pg_control. WriteControlFile() initializes pg_control
4211 : * given a preloaded buffer, ReadControlFile() loads the buffer from
4212 : * the pg_control file (during postmaster or standalone-backend startup),
4213 : * and UpdateControlFile() rewrites pg_control after we modify xlog state.
4214 : * InitControlFile() fills the buffer with initial values.
4215 : *
4216 : * For simplicity, WriteControlFile() initializes the fields of pg_control
4217 : * that are related to checking backend/database compatibility, and
4218 : * ReadControlFile() verifies they are correct. We could split out the
4219 : * I/O and compatibility-check functions, but there seems no need currently.
4220 : */
4221 :
4222 : static void
4223 100 : InitControlFile(uint64 sysidentifier, uint32 data_checksum_version)
4224 : {
4225 : char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4226 :
4227 : /*
4228 : * Generate a random nonce. This is used for authentication requests that
4229 : * will fail because the user does not exist. The nonce is used to create
4230 : * a genuine-looking password challenge for the non-existent user, in lieu
4231 : * of an actual stored password.
4232 : */
4233 100 : if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
4234 0 : ereport(PANIC,
4235 : (errcode(ERRCODE_INTERNAL_ERROR),
4236 : errmsg("could not generate secret authorization token")));
4237 :
4238 100 : memset(ControlFile, 0, sizeof(ControlFileData));
4239 : /* Initialize pg_control status fields */
4240 100 : ControlFile->system_identifier = sysidentifier;
4241 100 : memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4242 100 : ControlFile->state = DB_SHUTDOWNED;
4243 100 : ControlFile->unloggedLSN = FirstNormalUnloggedLSN;
4244 :
4245 : /* Set important parameter values for use when replaying WAL */
4246 100 : ControlFile->MaxConnections = MaxConnections;
4247 100 : ControlFile->max_worker_processes = max_worker_processes;
4248 100 : ControlFile->max_wal_senders = max_wal_senders;
4249 100 : ControlFile->max_prepared_xacts = max_prepared_xacts;
4250 100 : ControlFile->max_locks_per_xact = max_locks_per_xact;
4251 100 : ControlFile->wal_level = wal_level;
4252 100 : ControlFile->wal_log_hints = wal_log_hints;
4253 100 : ControlFile->track_commit_timestamp = track_commit_timestamp;
4254 100 : ControlFile->data_checksum_version = data_checksum_version;
4255 100 : }
4256 :
4257 : static void
4258 100 : WriteControlFile(void)
4259 : {
4260 : int fd;
4261 : char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4262 :
4263 : /*
4264 : * Initialize version and compatibility-check fields
4265 : */
4266 100 : ControlFile->pg_control_version = PG_CONTROL_VERSION;
4267 100 : ControlFile->catalog_version_no = CATALOG_VERSION_NO;
4268 :
4269 100 : ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4270 100 : ControlFile->floatFormat = FLOATFORMAT_VALUE;
4271 :
4272 100 : ControlFile->blcksz = BLCKSZ;
4273 100 : ControlFile->relseg_size = RELSEG_SIZE;
4274 100 : ControlFile->xlog_blcksz = XLOG_BLCKSZ;
4275 100 : ControlFile->xlog_seg_size = wal_segment_size;
4276 :
4277 100 : ControlFile->nameDataLen = NAMEDATALEN;
4278 100 : ControlFile->indexMaxKeys = INDEX_MAX_KEYS;
4279 :
4280 100 : ControlFile->toast_max_chunk_size = TOAST_MAX_CHUNK_SIZE;
4281 100 : ControlFile->loblksize = LOBLKSIZE;
4282 :
4283 100 : ControlFile->float8ByVal = true; /* vestigial */
4284 :
4285 : /*
4286 : * Initialize the default 'char' signedness.
4287 : *
4288 : * The signedness of the char type is implementation-defined. For instance
4289 : * on x86 architecture CPUs, the char data type is typically treated as
4290 : * signed by default, whereas on aarch architecture CPUs, it is typically
4291 : * treated as unsigned by default. In v17 or earlier, we accidentally let
4292 : * C implementation signedness affect persistent data. This led to
4293 : * inconsistent results when comparing char data across different
4294 : * platforms.
4295 : *
4296 : * This flag can be used as a hint to ensure consistent behavior for
4297 : * pre-v18 data files that store data sorted by the 'char' type on disk,
4298 : * especially in cross-platform replication scenarios.
4299 : *
4300 : * Newly created database clusters unconditionally set the default char
4301 : * signedness to true. pg_upgrade changes this flag for clusters that were
4302 : * initialized on signedness=false platforms. As a result,
4303 : * signedness=false setting will become rare over time. If we had known
4304 : * about this problem during the last development cycle that forced initdb
4305 : * (v8.3), we would have made all clusters signed or all clusters
4306 : * unsigned. Making pg_upgrade the only source of signedness=false will
4307 : * cause the population of database clusters to converge toward that
4308 : * retrospective ideal.
4309 : */
4310 100 : ControlFile->default_char_signedness = true;
4311 :
4312 : /* Contents are protected with a CRC */
4313 100 : INIT_CRC32C(ControlFile->crc);
4314 100 : COMP_CRC32C(ControlFile->crc,
4315 : ControlFile,
4316 : offsetof(ControlFileData, crc));
4317 100 : FIN_CRC32C(ControlFile->crc);
4318 :
4319 : /*
4320 : * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4321 : * the excess over sizeof(ControlFileData). This reduces the odds of
4322 : * premature-EOF errors when reading pg_control. We'll still fail when we
4323 : * check the contents of the file, but hopefully with a more specific
4324 : * error than "couldn't read pg_control".
4325 : */
4326 100 : memset(buffer, 0, PG_CONTROL_FILE_SIZE);
4327 100 : memcpy(buffer, ControlFile, sizeof(ControlFileData));
4328 :
4329 100 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4330 : O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
4331 100 : if (fd < 0)
4332 0 : ereport(PANIC,
4333 : (errcode_for_file_access(),
4334 : errmsg("could not create file \"%s\": %m",
4335 : XLOG_CONTROL_FILE)));
4336 :
4337 100 : errno = 0;
4338 100 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
4339 100 : if (write(fd, buffer, PG_CONTROL_FILE_SIZE) != PG_CONTROL_FILE_SIZE)
4340 : {
4341 : /* if write didn't set errno, assume problem is no disk space */
4342 0 : if (errno == 0)
4343 0 : errno = ENOSPC;
4344 0 : ereport(PANIC,
4345 : (errcode_for_file_access(),
4346 : errmsg("could not write to file \"%s\": %m",
4347 : XLOG_CONTROL_FILE)));
4348 : }
4349 100 : pgstat_report_wait_end();
4350 :
4351 100 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
4352 100 : if (pg_fsync(fd) != 0)
4353 0 : ereport(PANIC,
4354 : (errcode_for_file_access(),
4355 : errmsg("could not fsync file \"%s\": %m",
4356 : XLOG_CONTROL_FILE)));
4357 100 : pgstat_report_wait_end();
4358 :
4359 100 : if (close(fd) != 0)
4360 0 : ereport(PANIC,
4361 : (errcode_for_file_access(),
4362 : errmsg("could not close file \"%s\": %m",
4363 : XLOG_CONTROL_FILE)));
4364 100 : }
4365 :
4366 : static void
4367 2014 : ReadControlFile(void)
4368 : {
4369 : pg_crc32c crc;
4370 : int fd;
4371 : char wal_segsz_str[20];
4372 : int r;
4373 :
4374 : /*
4375 : * Read data...
4376 : */
4377 2014 : fd = BasicOpenFile(XLOG_CONTROL_FILE,
4378 : O_RDWR | PG_BINARY);
4379 2014 : if (fd < 0)
4380 0 : ereport(PANIC,
4381 : (errcode_for_file_access(),
4382 : errmsg("could not open file \"%s\": %m",
4383 : XLOG_CONTROL_FILE)));
4384 :
4385 2014 : pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
4386 2014 : r = read(fd, ControlFile, sizeof(ControlFileData));
4387 2014 : if (r != sizeof(ControlFileData))
4388 : {
4389 0 : if (r < 0)
4390 0 : ereport(PANIC,
4391 : (errcode_for_file_access(),
4392 : errmsg("could not read file \"%s\": %m",
4393 : XLOG_CONTROL_FILE)));
4394 : else
4395 0 : ereport(PANIC,
4396 : (errcode(ERRCODE_DATA_CORRUPTED),
4397 : errmsg("could not read file \"%s\": read %d of %zu",
4398 : XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4399 : }
4400 2014 : pgstat_report_wait_end();
4401 :
4402 2014 : close(fd);
4403 :
4404 : /*
4405 : * Check for expected pg_control format version. If this is wrong, the
4406 : * CRC check will likely fail because we'll be checking the wrong number
4407 : * of bytes. Complaining about wrong version will probably be more
4408 : * enlightening than complaining about wrong CRC.
4409 : */
4410 :
4411 2014 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION && ControlFile->pg_control_version % 65536 == 0 && ControlFile->pg_control_version / 65536 != 0)
4412 0 : ereport(FATAL,
4413 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4414 : errmsg("database files are incompatible with server"),
4415 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4416 : " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4417 : ControlFile->pg_control_version, ControlFile->pg_control_version,
4418 : PG_CONTROL_VERSION, PG_CONTROL_VERSION),
4419 : errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4420 :
4421 2014 : if (ControlFile->pg_control_version != PG_CONTROL_VERSION)
4422 0 : ereport(FATAL,
4423 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4424 : errmsg("database files are incompatible with server"),
4425 : errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4426 : " but the server was compiled with PG_CONTROL_VERSION %d.",
4427 : ControlFile->pg_control_version, PG_CONTROL_VERSION),
4428 : errhint("It looks like you need to initdb.")));
4429 :
4430 : /* Now check the CRC. */
4431 2014 : INIT_CRC32C(crc);
4432 2014 : COMP_CRC32C(crc,
4433 : ControlFile,
4434 : offsetof(ControlFileData, crc));
4435 2014 : FIN_CRC32C(crc);
4436 :
4437 2014 : if (!EQ_CRC32C(crc, ControlFile->crc))
4438 0 : ereport(FATAL,
4439 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4440 : errmsg("incorrect checksum in control file")));
4441 :
4442 : /*
4443 : * Do compatibility checking immediately. If the database isn't
4444 : * compatible with the backend executable, we want to abort before we can
4445 : * possibly do any damage.
4446 : */
4447 2014 : if (ControlFile->catalog_version_no != CATALOG_VERSION_NO)
4448 0 : ereport(FATAL,
4449 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4450 : errmsg("database files are incompatible with server"),
4451 : /* translator: %s is a variable name and %d is its value */
4452 : errdetail("The database cluster was initialized with %s %d,"
4453 : " but the server was compiled with %s %d.",
4454 : "CATALOG_VERSION_NO", ControlFile->catalog_version_no,
4455 : "CATALOG_VERSION_NO", CATALOG_VERSION_NO),
4456 : errhint("It looks like you need to initdb.")));
4457 2014 : if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
4458 0 : ereport(FATAL,
4459 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4460 : errmsg("database files are incompatible with server"),
4461 : /* translator: %s is a variable name and %d is its value */
4462 : errdetail("The database cluster was initialized with %s %d,"
4463 : " but the server was compiled with %s %d.",
4464 : "MAXALIGN", ControlFile->maxAlign,
4465 : "MAXALIGN", MAXIMUM_ALIGNOF),
4466 : errhint("It looks like you need to initdb.")));
4467 2014 : if (ControlFile->floatFormat != FLOATFORMAT_VALUE)
4468 0 : ereport(FATAL,
4469 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4470 : errmsg("database files are incompatible with server"),
4471 : errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4472 : errhint("It looks like you need to initdb.")));
4473 2014 : if (ControlFile->blcksz != BLCKSZ)
4474 0 : ereport(FATAL,
4475 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4476 : errmsg("database files are incompatible with server"),
4477 : /* translator: %s is a variable name and %d is its value */
4478 : errdetail("The database cluster was initialized with %s %d,"
4479 : " but the server was compiled with %s %d.",
4480 : "BLCKSZ", ControlFile->blcksz,
4481 : "BLCKSZ", BLCKSZ),
4482 : errhint("It looks like you need to recompile or initdb.")));
4483 2014 : if (ControlFile->relseg_size != RELSEG_SIZE)
4484 0 : ereport(FATAL,
4485 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4486 : errmsg("database files are incompatible with server"),
4487 : /* translator: %s is a variable name and %d is its value */
4488 : errdetail("The database cluster was initialized with %s %d,"
4489 : " but the server was compiled with %s %d.",
4490 : "RELSEG_SIZE", ControlFile->relseg_size,
4491 : "RELSEG_SIZE", RELSEG_SIZE),
4492 : errhint("It looks like you need to recompile or initdb.")));
4493 2014 : if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
4494 0 : ereport(FATAL,
4495 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4496 : errmsg("database files are incompatible with server"),
4497 : /* translator: %s is a variable name and %d is its value */
4498 : errdetail("The database cluster was initialized with %s %d,"
4499 : " but the server was compiled with %s %d.",
4500 : "XLOG_BLCKSZ", ControlFile->xlog_blcksz,
4501 : "XLOG_BLCKSZ", XLOG_BLCKSZ),
4502 : errhint("It looks like you need to recompile or initdb.")));
4503 2014 : if (ControlFile->nameDataLen != NAMEDATALEN)
4504 0 : ereport(FATAL,
4505 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4506 : errmsg("database files are incompatible with server"),
4507 : /* translator: %s is a variable name and %d is its value */
4508 : errdetail("The database cluster was initialized with %s %d,"
4509 : " but the server was compiled with %s %d.",
4510 : "NAMEDATALEN", ControlFile->nameDataLen,
4511 : "NAMEDATALEN", NAMEDATALEN),
4512 : errhint("It looks like you need to recompile or initdb.")));
4513 2014 : if (ControlFile->indexMaxKeys != INDEX_MAX_KEYS)
4514 0 : ereport(FATAL,
4515 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4516 : errmsg("database files are incompatible with server"),
4517 : /* translator: %s is a variable name and %d is its value */
4518 : errdetail("The database cluster was initialized with %s %d,"
4519 : " but the server was compiled with %s %d.",
4520 : "INDEX_MAX_KEYS", ControlFile->indexMaxKeys,
4521 : "INDEX_MAX_KEYS", INDEX_MAX_KEYS),
4522 : errhint("It looks like you need to recompile or initdb.")));
4523 2014 : if (ControlFile->toast_max_chunk_size != TOAST_MAX_CHUNK_SIZE)
4524 0 : ereport(FATAL,
4525 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4526 : errmsg("database files are incompatible with server"),
4527 : /* translator: %s is a variable name and %d is its value */
4528 : errdetail("The database cluster was initialized with %s %d,"
4529 : " but the server was compiled with %s %d.",
4530 : "TOAST_MAX_CHUNK_SIZE", ControlFile->toast_max_chunk_size,
4531 : "TOAST_MAX_CHUNK_SIZE", (int) TOAST_MAX_CHUNK_SIZE),
4532 : errhint("It looks like you need to recompile or initdb.")));
4533 2014 : if (ControlFile->loblksize != LOBLKSIZE)
4534 0 : ereport(FATAL,
4535 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4536 : errmsg("database files are incompatible with server"),
4537 : /* translator: %s is a variable name and %d is its value */
4538 : errdetail("The database cluster was initialized with %s %d,"
4539 : " but the server was compiled with %s %d.",
4540 : "LOBLKSIZE", ControlFile->loblksize,
4541 : "LOBLKSIZE", (int) LOBLKSIZE),
4542 : errhint("It looks like you need to recompile or initdb.")));
4543 :
4544 : Assert(ControlFile->float8ByVal); /* vestigial, not worth an error msg */
4545 :
4546 2014 : wal_segment_size = ControlFile->xlog_seg_size;
4547 :
4548 2014 : if (!IsValidWalSegSize(wal_segment_size))
4549 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4550 : errmsg_plural("invalid WAL segment size in control file (%d byte)",
4551 : "invalid WAL segment size in control file (%d bytes)",
4552 : wal_segment_size,
4553 : wal_segment_size),
4554 : errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.")));
4555 :
4556 2014 : snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4557 2014 : SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4558 : PGC_S_DYNAMIC_DEFAULT);
4559 :
4560 : /* check and update variables dependent on wal_segment_size */
4561 2014 : if (ConvertToXSegs(min_wal_size_mb, wal_segment_size) < 2)
4562 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4563 : /* translator: both %s are GUC names */
4564 : errmsg("\"%s\" must be at least twice \"%s\"",
4565 : "min_wal_size", "wal_segment_size")));
4566 :
4567 2014 : if (ConvertToXSegs(max_wal_size_mb, wal_segment_size) < 2)
4568 0 : ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4569 : /* translator: both %s are GUC names */
4570 : errmsg("\"%s\" must be at least twice \"%s\"",
4571 : "max_wal_size", "wal_segment_size")));
4572 :
4573 2014 : UsableBytesInSegment =
4574 2014 : (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4575 : (SizeOfXLogLongPHD - SizeOfXLogShortPHD);
4576 :
4577 2014 : CalculateCheckpointSegments();
4578 :
4579 : /* Make the initdb settings visible as GUC variables, too */
4580 2014 : SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4581 : PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT);
4582 2014 : }
4583 :
4584 : /*
4585 : * Utility wrapper to update the control file. Note that the control
4586 : * file gets flushed.
4587 : */
4588 : static void
4589 18130 : UpdateControlFile(void)
4590 : {
4591 18130 : update_controlfile(DataDir, ControlFile, true);
4592 18130 : }
4593 :
4594 : /*
4595 : * Returns the unique system identifier from control file.
4596 : */
4597 : uint64
4598 2808 : GetSystemIdentifier(void)
4599 : {
4600 : Assert(ControlFile != NULL);
4601 2808 : return ControlFile->system_identifier;
4602 : }
4603 :
4604 : /*
4605 : * Returns the random nonce from control file.
4606 : */
4607 : char *
4608 2 : GetMockAuthenticationNonce(void)
4609 : {
4610 : Assert(ControlFile != NULL);
4611 2 : return ControlFile->mock_authentication_nonce;
4612 : }
4613 :
4614 : /*
4615 : * Are checksums enabled for data pages?
4616 : */
4617 : bool
4618 20036568 : DataChecksumsEnabled(void)
4619 : {
4620 : Assert(ControlFile != NULL);
4621 20036568 : return (ControlFile->data_checksum_version > 0);
4622 : }
4623 :
4624 : /*
4625 : * Return true if the cluster was initialized on a platform where the
4626 : * default signedness of char is "signed". This function exists for code
4627 : * that deals with pre-v18 data files that store data sorted by the 'char'
4628 : * type on disk (e.g., GIN and GiST indexes). See the comments in
4629 : * WriteControlFile() for details.
4630 : */
4631 : bool
4632 6 : GetDefaultCharSignedness(void)
4633 : {
4634 6 : return ControlFile->default_char_signedness;
4635 : }
4636 :
4637 : /*
4638 : * Returns a fake LSN for unlogged relations.
4639 : *
4640 : * Each call generates an LSN that is greater than any previous value
4641 : * returned. The current counter value is saved and restored across clean
4642 : * shutdowns, but like unlogged relations, does not survive a crash. This can
4643 : * be used in lieu of real LSN values returned by XLogInsert, if you need an
4644 : * LSN-like increasing sequence of numbers without writing any WAL.
4645 : */
4646 : XLogRecPtr
4647 66 : GetFakeLSNForUnloggedRel(void)
4648 : {
4649 66 : return pg_atomic_fetch_add_u64(&XLogCtl->unloggedLSN, 1);
4650 : }
4651 :
4652 : /*
4653 : * Auto-tune the number of XLOG buffers.
4654 : *
4655 : * The preferred setting for wal_buffers is about 3% of shared_buffers, with
4656 : * a maximum of one XLOG segment (there is little reason to think that more
4657 : * is helpful, at least so long as we force an fsync when switching log files)
4658 : * and a minimum of 8 blocks (which was the default value prior to PostgreSQL
4659 : * 9.1, when auto-tuning was added).
4660 : *
4661 : * This should not be called until NBuffers has received its final value.
4662 : */
4663 : static int
4664 2194 : XLOGChooseNumBuffers(void)
4665 : {
4666 : int xbuffers;
4667 :
4668 2194 : xbuffers = NBuffers / 32;
4669 2194 : if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4670 48 : xbuffers = (wal_segment_size / XLOG_BLCKSZ);
4671 2194 : if (xbuffers < 8)
4672 854 : xbuffers = 8;
4673 2194 : return xbuffers;
4674 : }
4675 :
4676 : /*
4677 : * GUC check_hook for wal_buffers
4678 : */
4679 : bool
4680 4466 : check_wal_buffers(int *newval, void **extra, GucSource source)
4681 : {
4682 : /*
4683 : * -1 indicates a request for auto-tune.
4684 : */
4685 4466 : if (*newval == -1)
4686 : {
4687 : /*
4688 : * If we haven't yet changed the boot_val default of -1, just let it
4689 : * be. We'll fix it when XLOGShmemSize is called.
4690 : */
4691 2272 : if (XLOGbuffers == -1)
4692 2272 : return true;
4693 :
4694 : /* Otherwise, substitute the auto-tune value */
4695 0 : *newval = XLOGChooseNumBuffers();
4696 : }
4697 :
4698 : /*
4699 : * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4700 : * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4701 : * the case, we just silently treat such values as a request for the
4702 : * minimum. (We could throw an error instead, but that doesn't seem very
4703 : * helpful.)
4704 : */
4705 2194 : if (*newval < 4)
4706 0 : *newval = 4;
4707 :
4708 2194 : return true;
4709 : }
4710 :
4711 : /*
4712 : * GUC check_hook for wal_consistency_checking
4713 : */
4714 : bool
4715 4076 : check_wal_consistency_checking(char **newval, void **extra, GucSource source)
4716 : {
4717 : char *rawstring;
4718 : List *elemlist;
4719 : ListCell *l;
4720 : bool newwalconsistency[RM_MAX_ID + 1];
4721 :
4722 : /* Initialize the array */
4723 134508 : MemSet(newwalconsistency, 0, (RM_MAX_ID + 1) * sizeof(bool));
4724 :
4725 : /* Need a modifiable copy of string */
4726 4076 : rawstring = pstrdup(*newval);
4727 :
4728 : /* Parse string into list of identifiers */
4729 4076 : if (!SplitIdentifierString(rawstring, ',', &elemlist))
4730 : {
4731 : /* syntax error in list */
4732 0 : GUC_check_errdetail("List syntax is invalid.");
4733 0 : pfree(rawstring);
4734 0 : list_free(elemlist);
4735 0 : return false;
4736 : }
4737 :
4738 4984 : foreach(l, elemlist)
4739 : {
4740 908 : char *tok = (char *) lfirst(l);
4741 : int rmid;
4742 :
4743 : /* Check for 'all'. */
4744 908 : if (pg_strcasecmp(tok, "all") == 0)
4745 : {
4746 232328 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4747 231424 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL)
4748 9040 : newwalconsistency[rmid] = true;
4749 : }
4750 : else
4751 : {
4752 : /* Check if the token matches any known resource manager. */
4753 4 : bool found = false;
4754 :
4755 72 : for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4756 : {
4757 108 : if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL &&
4758 36 : pg_strcasecmp(tok, GetRmgr(rmid).rm_name) == 0)
4759 : {
4760 4 : newwalconsistency[rmid] = true;
4761 4 : found = true;
4762 4 : break;
4763 : }
4764 : }
4765 4 : if (!found)
4766 : {
4767 : /*
4768 : * During startup, it might be a not-yet-loaded custom
4769 : * resource manager. Defer checking until
4770 : * InitializeWalConsistencyChecking().
4771 : */
4772 0 : if (!process_shared_preload_libraries_done)
4773 : {
4774 0 : check_wal_consistency_checking_deferred = true;
4775 : }
4776 : else
4777 : {
4778 0 : GUC_check_errdetail("Unrecognized key word: \"%s\".", tok);
4779 0 : pfree(rawstring);
4780 0 : list_free(elemlist);
4781 0 : return false;
4782 : }
4783 : }
4784 : }
4785 : }
4786 :
4787 4076 : pfree(rawstring);
4788 4076 : list_free(elemlist);
4789 :
4790 : /* assign new value */
4791 4076 : *extra = guc_malloc(LOG, (RM_MAX_ID + 1) * sizeof(bool));
4792 4076 : if (!*extra)
4793 0 : return false;
4794 4076 : memcpy(*extra, newwalconsistency, (RM_MAX_ID + 1) * sizeof(bool));
4795 4076 : return true;
4796 : }
4797 :
4798 : /*
4799 : * GUC assign_hook for wal_consistency_checking
4800 : */
4801 : void
4802 4074 : assign_wal_consistency_checking(const char *newval, void *extra)
4803 : {
4804 : /*
4805 : * If some checks were deferred, it's possible that the checks will fail
4806 : * later during InitializeWalConsistencyChecking(). But in that case, the
4807 : * postmaster will exit anyway, so it's safe to proceed with the
4808 : * assignment.
4809 : *
4810 : * Any built-in resource managers specified are assigned immediately,
4811 : * which affects WAL created before shared_preload_libraries are
4812 : * processed. Any custom resource managers specified won't be assigned
4813 : * until after shared_preload_libraries are processed, but that's OK
4814 : * because WAL for a custom resource manager can't be written before the
4815 : * module is loaded anyway.
4816 : */
4817 4074 : wal_consistency_checking = extra;
4818 4074 : }
4819 :
4820 : /*
4821 : * InitializeWalConsistencyChecking: run after loading custom resource managers
4822 : *
4823 : * If any unknown resource managers were specified in the
4824 : * wal_consistency_checking GUC, processing was deferred. Now that
4825 : * shared_preload_libraries have been loaded, process wal_consistency_checking
4826 : * again.
4827 : */
4828 : void
4829 1894 : InitializeWalConsistencyChecking(void)
4830 : {
4831 : Assert(process_shared_preload_libraries_done);
4832 :
4833 1894 : if (check_wal_consistency_checking_deferred)
4834 : {
4835 : struct config_generic *guc;
4836 :
4837 0 : guc = find_option("wal_consistency_checking", false, false, ERROR);
4838 :
4839 0 : check_wal_consistency_checking_deferred = false;
4840 :
4841 0 : set_config_option_ext("wal_consistency_checking",
4842 : wal_consistency_checking_string,
4843 : guc->scontext, guc->source, guc->srole,
4844 : GUC_ACTION_SET, true, ERROR, false);
4845 :
4846 : /* checking should not be deferred again */
4847 : Assert(!check_wal_consistency_checking_deferred);
4848 : }
4849 1894 : }
4850 :
4851 : /*
4852 : * GUC show_hook for archive_command
4853 : */
4854 : const char *
4855 3458 : show_archive_command(void)
4856 : {
4857 3458 : if (XLogArchivingActive())
4858 4 : return XLogArchiveCommand;
4859 : else
4860 3454 : return "(disabled)";
4861 : }
4862 :
4863 : /*
4864 : * GUC show_hook for in_hot_standby
4865 : */
4866 : const char *
4867 30316 : show_in_hot_standby(void)
4868 : {
4869 : /*
4870 : * We display the actual state based on shared memory, so that this GUC
4871 : * reports up-to-date state if examined intra-query. The underlying
4872 : * variable (in_hot_standby_guc) changes only when we transmit a new value
4873 : * to the client.
4874 : */
4875 30316 : return RecoveryInProgress() ? "on" : "off";
4876 : }
4877 :
4878 : /*
4879 : * Read the control file, set respective GUCs.
4880 : *
4881 : * This is to be called during startup, including a crash recovery cycle,
4882 : * unless in bootstrap mode, where no control file yet exists. As there's no
4883 : * usable shared memory yet (its sizing can depend on the contents of the
4884 : * control file!), first store the contents in local memory. XLOGShmemInit()
4885 : * will then copy it to shared memory later.
4886 : *
4887 : * reset just controls whether previous contents are to be expected (in the
4888 : * reset case, there's a dangling pointer into old shared memory), or not.
4889 : */
4890 : void
4891 1914 : LocalProcessControlFile(bool reset)
4892 : {
4893 : Assert(reset || ControlFile == NULL);
4894 1914 : ControlFile = palloc(sizeof(ControlFileData));
4895 1914 : ReadControlFile();
4896 1914 : }
4897 :
4898 : /*
4899 : * Get the wal_level from the control file. For a standby, this value should be
4900 : * considered as its active wal_level, because it may be different from what
4901 : * was originally configured on standby.
4902 : */
4903 : WalLevel
4904 138 : GetActiveWalLevelOnStandby(void)
4905 : {
4906 138 : return ControlFile->wal_level;
4907 : }
4908 :
4909 : /*
4910 : * Initialization of shared memory for XLOG
4911 : */
4912 : Size
4913 6294 : XLOGShmemSize(void)
4914 : {
4915 : Size size;
4916 :
4917 : /*
4918 : * If the value of wal_buffers is -1, use the preferred auto-tune value.
4919 : * This isn't an amazingly clean place to do this, but we must wait till
4920 : * NBuffers has received its final value, and must do it before using the
4921 : * value of XLOGbuffers to do anything important.
4922 : *
4923 : * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4924 : * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4925 : * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4926 : * the matter with PGC_S_OVERRIDE.
4927 : */
4928 6294 : if (XLOGbuffers == -1)
4929 : {
4930 : char buf[32];
4931 :
4932 2194 : snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
4933 2194 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4934 : PGC_S_DYNAMIC_DEFAULT);
4935 2194 : if (XLOGbuffers == -1) /* failed to apply it? */
4936 0 : SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4937 : PGC_S_OVERRIDE);
4938 : }
4939 : Assert(XLOGbuffers > 0);
4940 :
4941 : /* XLogCtl */
4942 6294 : size = sizeof(XLogCtlData);
4943 :
4944 : /* WAL insertion locks, plus alignment */
4945 6294 : size = add_size(size, mul_size(sizeof(WALInsertLockPadded), NUM_XLOGINSERT_LOCKS + 1));
4946 : /* xlblocks array */
4947 6294 : size = add_size(size, mul_size(sizeof(pg_atomic_uint64), XLOGbuffers));
4948 : /* extra alignment padding for XLOG I/O buffers */
4949 6294 : size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4950 : /* and the buffers themselves */
4951 6294 : size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4952 :
4953 : /*
4954 : * Note: we don't count ControlFileData, it comes out of the "slop factor"
4955 : * added by CreateSharedMemoryAndSemaphores. This lets us use this
4956 : * routine again below to compute the actual allocation size.
4957 : */
4958 :
4959 6294 : return size;
4960 : }
4961 :
4962 : void
4963 2198 : XLOGShmemInit(void)
4964 : {
4965 : bool foundCFile,
4966 : foundXLog;
4967 : char *allocptr;
4968 : int i;
4969 : ControlFileData *localControlFile;
4970 :
4971 : #ifdef WAL_DEBUG
4972 :
4973 : /*
4974 : * Create a memory context for WAL debugging that's exempt from the normal
4975 : * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4976 : * an allocation fails, but wal_debug is not for production use anyway.
4977 : */
4978 : if (walDebugCxt == NULL)
4979 : {
4980 : walDebugCxt = AllocSetContextCreate(TopMemoryContext,
4981 : "WAL Debug",
4982 : ALLOCSET_DEFAULT_SIZES);
4983 : MemoryContextAllowInCriticalSection(walDebugCxt, true);
4984 : }
4985 : #endif
4986 :
4987 :
4988 2198 : XLogCtl = (XLogCtlData *)
4989 2198 : ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4990 :
4991 2198 : localControlFile = ControlFile;
4992 2198 : ControlFile = (ControlFileData *)
4993 2198 : ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4994 :
4995 2198 : if (foundCFile || foundXLog)
4996 : {
4997 : /* both should be present or neither */
4998 : Assert(foundCFile && foundXLog);
4999 :
5000 : /* Initialize local copy of WALInsertLocks */
5001 0 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks;
5002 :
5003 0 : if (localControlFile)
5004 0 : pfree(localControlFile);
5005 0 : return;
5006 : }
5007 2198 : memset(XLogCtl, 0, sizeof(XLogCtlData));
5008 :
5009 : /*
5010 : * Already have read control file locally, unless in bootstrap mode. Move
5011 : * contents into shared memory.
5012 : */
5013 2198 : if (localControlFile)
5014 : {
5015 1898 : memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
5016 1898 : pfree(localControlFile);
5017 : }
5018 :
5019 : /*
5020 : * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
5021 : * multiple of the alignment for same, so no extra alignment padding is
5022 : * needed here.
5023 : */
5024 2198 : allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
5025 2198 : XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
5026 2198 : allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
5027 :
5028 627602 : for (i = 0; i < XLOGbuffers; i++)
5029 : {
5030 625404 : pg_atomic_init_u64(&XLogCtl->xlblocks[i], InvalidXLogRecPtr);
5031 : }
5032 :
5033 : /* WAL insertion locks. Ensure they're aligned to the full padded size */
5034 2198 : allocptr += sizeof(WALInsertLockPadded) -
5035 2198 : ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
5036 2198 : WALInsertLocks = XLogCtl->Insert.WALInsertLocks =
5037 : (WALInsertLockPadded *) allocptr;
5038 2198 : allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
5039 :
5040 19782 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
5041 : {
5042 17584 : LWLockInitialize(&WALInsertLocks[i].l.lock, LWTRANCHE_WAL_INSERT);
5043 17584 : pg_atomic_init_u64(&WALInsertLocks[i].l.insertingAt, InvalidXLogRecPtr);
5044 17584 : WALInsertLocks[i].l.lastImportantAt = InvalidXLogRecPtr;
5045 : }
5046 :
5047 : /*
5048 : * Align the start of the page buffers to a full xlog block size boundary.
5049 : * This simplifies some calculations in XLOG insertion. It is also
5050 : * required for O_DIRECT.
5051 : */
5052 2198 : allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
5053 2198 : XLogCtl->pages = allocptr;
5054 2198 : memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5055 :
5056 : /*
5057 : * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5058 : * in additional info.)
5059 : */
5060 2198 : XLogCtl->XLogCacheBlck = XLOGbuffers - 1;
5061 2198 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5062 2198 : XLogCtl->InstallXLogFileSegmentActive = false;
5063 2198 : XLogCtl->WalWriterSleeping = false;
5064 :
5065 2198 : SpinLockInit(&XLogCtl->Insert.insertpos_lck);
5066 2198 : SpinLockInit(&XLogCtl->info_lck);
5067 2198 : pg_atomic_init_u64(&XLogCtl->logInsertResult, InvalidXLogRecPtr);
5068 2198 : pg_atomic_init_u64(&XLogCtl->logWriteResult, InvalidXLogRecPtr);
5069 2198 : pg_atomic_init_u64(&XLogCtl->logFlushResult, InvalidXLogRecPtr);
5070 2198 : pg_atomic_init_u64(&XLogCtl->unloggedLSN, InvalidXLogRecPtr);
5071 : }
5072 :
5073 : /*
5074 : * This func must be called ONCE on system install. It creates pg_control
5075 : * and the initial XLOG segment.
5076 : */
5077 : void
5078 100 : BootStrapXLOG(uint32 data_checksum_version)
5079 : {
5080 : CheckPoint checkPoint;
5081 : char *buffer;
5082 : XLogPageHeader page;
5083 : XLogLongPageHeader longpage;
5084 : XLogRecord *record;
5085 : char *recptr;
5086 : uint64 sysidentifier;
5087 : struct timeval tv;
5088 : pg_crc32c crc;
5089 :
5090 : /* allow ordinary WAL segment creation, like StartupXLOG() would */
5091 100 : SetInstallXLogFileSegmentActive();
5092 :
5093 : /*
5094 : * Select a hopefully-unique system identifier code for this installation.
5095 : * We use the result of gettimeofday(), including the fractional seconds
5096 : * field, as being about as unique as we can easily get. (Think not to
5097 : * use random(), since it hasn't been seeded and there's no portable way
5098 : * to seed it other than the system clock value...) The upper half of the
5099 : * uint64 value is just the tv_sec part, while the lower half contains the
5100 : * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5101 : * PID for a little extra uniqueness. A person knowing this encoding can
5102 : * determine the initialization time of the installation, which could
5103 : * perhaps be useful sometimes.
5104 : */
5105 100 : gettimeofday(&tv, NULL);
5106 100 : sysidentifier = ((uint64) tv.tv_sec) << 32;
5107 100 : sysidentifier |= ((uint64) tv.tv_usec) << 12;
5108 100 : sysidentifier |= getpid() & 0xFFF;
5109 :
5110 : /* page buffer must be aligned suitably for O_DIRECT */
5111 100 : buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5112 100 : page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
5113 100 : memset(page, 0, XLOG_BLCKSZ);
5114 :
5115 : /*
5116 : * Set up information for the initial checkpoint record
5117 : *
5118 : * The initial checkpoint record is written to the beginning of the WAL
5119 : * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5120 : * used, so that we can use 0/0 to mean "before any valid WAL segment".
5121 : */
5122 100 : checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
5123 100 : checkPoint.ThisTimeLineID = BootstrapTimeLineID;
5124 100 : checkPoint.PrevTimeLineID = BootstrapTimeLineID;
5125 100 : checkPoint.fullPageWrites = fullPageWrites;
5126 100 : checkPoint.wal_level = wal_level;
5127 : checkPoint.nextXid =
5128 100 : FullTransactionIdFromEpochAndXid(0, FirstNormalTransactionId);
5129 100 : checkPoint.nextOid = FirstGenbkiObjectId;
5130 100 : checkPoint.nextMulti = FirstMultiXactId;
5131 100 : checkPoint.nextMultiOffset = 0;
5132 100 : checkPoint.oldestXid = FirstNormalTransactionId;
5133 100 : checkPoint.oldestXidDB = Template1DbOid;
5134 100 : checkPoint.oldestMulti = FirstMultiXactId;
5135 100 : checkPoint.oldestMultiDB = Template1DbOid;
5136 100 : checkPoint.oldestCommitTsXid = InvalidTransactionId;
5137 100 : checkPoint.newestCommitTsXid = InvalidTransactionId;
5138 100 : checkPoint.time = (pg_time_t) time(NULL);
5139 100 : checkPoint.oldestActiveXid = InvalidTransactionId;
5140 :
5141 100 : TransamVariables->nextXid = checkPoint.nextXid;
5142 100 : TransamVariables->nextOid = checkPoint.nextOid;
5143 100 : TransamVariables->oidCount = 0;
5144 100 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5145 100 : AdvanceOldestClogXid(checkPoint.oldestXid);
5146 100 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5147 100 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5148 100 : SetCommitTsLimit(InvalidTransactionId, InvalidTransactionId);
5149 :
5150 : /* Set up the XLOG page header */
5151 100 : page->xlp_magic = XLOG_PAGE_MAGIC;
5152 100 : page->xlp_info = XLP_LONG_HEADER;
5153 100 : page->xlp_tli = BootstrapTimeLineID;
5154 100 : page->xlp_pageaddr = wal_segment_size;
5155 100 : longpage = (XLogLongPageHeader) page;
5156 100 : longpage->xlp_sysid = sysidentifier;
5157 100 : longpage->xlp_seg_size = wal_segment_size;
5158 100 : longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5159 :
5160 : /* Insert the initial checkpoint record */
5161 100 : recptr = ((char *) page + SizeOfXLogLongPHD);
5162 100 : record = (XLogRecord *) recptr;
5163 100 : record->xl_prev = 0;
5164 100 : record->xl_xid = InvalidTransactionId;
5165 100 : record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5166 100 : record->xl_info = XLOG_CHECKPOINT_SHUTDOWN;
5167 100 : record->xl_rmid = RM_XLOG_ID;
5168 100 : recptr += SizeOfXLogRecord;
5169 : /* fill the XLogRecordDataHeaderShort struct */
5170 100 : *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5171 100 : *(recptr++) = sizeof(checkPoint);
5172 100 : memcpy(recptr, &checkPoint, sizeof(checkPoint));
5173 100 : recptr += sizeof(checkPoint);
5174 : Assert(recptr - (char *) record == record->xl_tot_len);
5175 :
5176 100 : INIT_CRC32C(crc);
5177 100 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5178 100 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5179 100 : FIN_CRC32C(crc);
5180 100 : record->xl_crc = crc;
5181 :
5182 : /* Create first XLOG segment file */
5183 100 : openLogTLI = BootstrapTimeLineID;
5184 100 : openLogFile = XLogFileInit(1, BootstrapTimeLineID);
5185 :
5186 : /*
5187 : * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5188 : * close the file again in a moment.
5189 : */
5190 :
5191 : /* Write the first page with the initial record */
5192 100 : errno = 0;
5193 100 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5194 100 : if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5195 : {
5196 : /* if write didn't set errno, assume problem is no disk space */
5197 0 : if (errno == 0)
5198 0 : errno = ENOSPC;
5199 0 : ereport(PANIC,
5200 : (errcode_for_file_access(),
5201 : errmsg("could not write bootstrap write-ahead log file: %m")));
5202 : }
5203 100 : pgstat_report_wait_end();
5204 :
5205 100 : pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5206 100 : if (pg_fsync(openLogFile) != 0)
5207 0 : ereport(PANIC,
5208 : (errcode_for_file_access(),
5209 : errmsg("could not fsync bootstrap write-ahead log file: %m")));
5210 100 : pgstat_report_wait_end();
5211 :
5212 100 : if (close(openLogFile) != 0)
5213 0 : ereport(PANIC,
5214 : (errcode_for_file_access(),
5215 : errmsg("could not close bootstrap write-ahead log file: %m")));
5216 :
5217 100 : openLogFile = -1;
5218 :
5219 : /* Now create pg_control */
5220 100 : InitControlFile(sysidentifier, data_checksum_version);
5221 100 : ControlFile->time = checkPoint.time;
5222 100 : ControlFile->checkPoint = checkPoint.redo;
5223 100 : ControlFile->checkPointCopy = checkPoint;
5224 :
5225 : /* some additional ControlFile fields are set in WriteControlFile() */
5226 100 : WriteControlFile();
5227 :
5228 : /* Bootstrap the commit log, too */
5229 100 : BootStrapCLOG();
5230 100 : BootStrapCommitTs();
5231 100 : BootStrapSUBTRANS();
5232 100 : BootStrapMultiXact();
5233 :
5234 100 : pfree(buffer);
5235 :
5236 : /*
5237 : * Force control file to be read - in contrast to normal processing we'd
5238 : * otherwise never run the checks and GUC related initializations therein.
5239 : */
5240 100 : ReadControlFile();
5241 100 : }
5242 :
5243 : static char *
5244 1694 : str_time(pg_time_t tnow, char *buf, size_t bufsize)
5245 : {
5246 1694 : pg_strftime(buf, bufsize,
5247 : "%Y-%m-%d %H:%M:%S %Z",
5248 1694 : pg_localtime(&tnow, log_timezone));
5249 :
5250 1694 : return buf;
5251 : }
5252 :
5253 : /*
5254 : * Initialize the first WAL segment on new timeline.
5255 : */
5256 : static void
5257 102 : XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
5258 : {
5259 : char xlogfname[MAXFNAMELEN];
5260 : XLogSegNo endLogSegNo;
5261 : XLogSegNo startLogSegNo;
5262 :
5263 : /* we always switch to a new timeline after archive recovery */
5264 : Assert(endTLI != newTLI);
5265 :
5266 : /*
5267 : * Update min recovery point one last time.
5268 : */
5269 102 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
5270 :
5271 : /*
5272 : * Calculate the last segment on the old timeline, and the first segment
5273 : * on the new timeline. If the switch happens in the middle of a segment,
5274 : * they are the same, but if the switch happens exactly at a segment
5275 : * boundary, startLogSegNo will be endLogSegNo + 1.
5276 : */
5277 102 : XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5278 102 : XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5279 :
5280 : /*
5281 : * Initialize the starting WAL segment for the new timeline. If the switch
5282 : * happens in the middle of a segment, copy data from the last WAL segment
5283 : * of the old timeline up to the switch point, to the starting WAL segment
5284 : * on the new timeline.
5285 : */
5286 102 : if (endLogSegNo == startLogSegNo)
5287 : {
5288 : /*
5289 : * Make a copy of the file on the new timeline.
5290 : *
5291 : * Writing WAL isn't allowed yet, so there are no locking
5292 : * considerations. But we should be just as tense as XLogFileInit to
5293 : * avoid emplacing a bogus file.
5294 : */
5295 84 : XLogFileCopy(newTLI, endLogSegNo, endTLI, endLogSegNo,
5296 84 : XLogSegmentOffset(endOfLog, wal_segment_size));
5297 : }
5298 : else
5299 : {
5300 : /*
5301 : * The switch happened at a segment boundary, so just create the next
5302 : * segment on the new timeline.
5303 : */
5304 : int fd;
5305 :
5306 18 : fd = XLogFileInit(startLogSegNo, newTLI);
5307 :
5308 18 : if (close(fd) != 0)
5309 : {
5310 0 : int save_errno = errno;
5311 :
5312 0 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
5313 0 : errno = save_errno;
5314 0 : ereport(ERROR,
5315 : (errcode_for_file_access(),
5316 : errmsg("could not close file \"%s\": %m", xlogfname)));
5317 : }
5318 : }
5319 :
5320 : /*
5321 : * Let's just make real sure there are not .ready or .done flags posted
5322 : * for the new segment.
5323 : */
5324 102 : XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
5325 102 : XLogArchiveCleanup(xlogfname);
5326 102 : }
5327 :
5328 : /*
5329 : * Perform cleanup actions at the conclusion of archive recovery.
5330 : */
5331 : static void
5332 102 : CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
5333 : TimeLineID newTLI)
5334 : {
5335 : /*
5336 : * Execute the recovery_end_command, if any.
5337 : */
5338 102 : if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
5339 4 : ExecuteRecoveryCommand(recoveryEndCommand,
5340 : "recovery_end_command",
5341 : true,
5342 : WAIT_EVENT_RECOVERY_END_COMMAND);
5343 :
5344 : /*
5345 : * We switched to a new timeline. Clean up segments on the old timeline.
5346 : *
5347 : * If there are any higher-numbered segments on the old timeline, remove
5348 : * them. They might contain valid WAL, but they might also be
5349 : * pre-allocated files containing garbage. In any case, they are not part
5350 : * of the new timeline's history so we don't need them.
5351 : */
5352 102 : RemoveNonParentXlogFiles(EndOfLog, newTLI);
5353 :
5354 : /*
5355 : * If the switch happened in the middle of a segment, what to do with the
5356 : * last, partial segment on the old timeline? If we don't archive it, and
5357 : * the server that created the WAL never archives it either (e.g. because
5358 : * it was hit by a meteor), it will never make it to the archive. That's
5359 : * OK from our point of view, because the new segment that we created with
5360 : * the new TLI contains all the WAL from the old timeline up to the switch
5361 : * point. But if you later try to do PITR to the "missing" WAL on the old
5362 : * timeline, recovery won't find it in the archive. It's physically
5363 : * present in the new file with new TLI, but recovery won't look there
5364 : * when it's recovering to the older timeline. On the other hand, if we
5365 : * archive the partial segment, and the original server on that timeline
5366 : * is still running and archives the completed version of the same segment
5367 : * later, it will fail. (We used to do that in 9.4 and below, and it
5368 : * caused such problems).
5369 : *
5370 : * As a compromise, we rename the last segment with the .partial suffix,
5371 : * and archive it. Archive recovery will never try to read .partial
5372 : * segments, so they will normally go unused. But in the odd PITR case,
5373 : * the administrator can copy them manually to the pg_wal directory
5374 : * (removing the suffix). They can be useful in debugging, too.
5375 : *
5376 : * If a .done or .ready file already exists for the old timeline, however,
5377 : * we had already determined that the segment is complete, so we can let
5378 : * it be archived normally. (In particular, if it was restored from the
5379 : * archive to begin with, it's expected to have a .done file).
5380 : */
5381 102 : if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
5382 : XLogArchivingActive())
5383 : {
5384 : char origfname[MAXFNAMELEN];
5385 : XLogSegNo endLogSegNo;
5386 :
5387 20 : XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
5388 20 : XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
5389 :
5390 20 : if (!XLogArchiveIsReadyOrDone(origfname))
5391 : {
5392 : char origpath[MAXPGPATH];
5393 : char partialfname[MAXFNAMELEN];
5394 : char partialpath[MAXPGPATH];
5395 :
5396 : /*
5397 : * If we're summarizing WAL, we can't rename the partial file
5398 : * until the summarizer finishes with it, else it will fail.
5399 : */
5400 12 : if (summarize_wal)
5401 2 : WaitForWalSummarization(EndOfLog);
5402 :
5403 12 : XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
5404 12 : snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
5405 12 : snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
5406 :
5407 : /*
5408 : * Make sure there's no .done or .ready file for the .partial
5409 : * file.
5410 : */
5411 12 : XLogArchiveCleanup(partialfname);
5412 :
5413 12 : durable_rename(origpath, partialpath, ERROR);
5414 12 : XLogArchiveNotify(partialfname);
5415 : }
5416 : }
5417 102 : }
5418 :
5419 : /*
5420 : * Check to see if required parameters are set high enough on this server
5421 : * for various aspects of recovery operation.
5422 : *
5423 : * Note that all the parameters which this function tests need to be
5424 : * listed in Administrator's Overview section in high-availability.sgml.
5425 : * If you change them, don't forget to update the list.
5426 : */
5427 : static void
5428 496 : CheckRequiredParameterValues(void)
5429 : {
5430 : /*
5431 : * For archive recovery, the WAL must be generated with at least 'replica'
5432 : * wal_level.
5433 : */
5434 496 : if (ArchiveRecoveryRequested && ControlFile->wal_level == WAL_LEVEL_MINIMAL)
5435 : {
5436 4 : ereport(FATAL,
5437 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5438 : errmsg("WAL was generated with \"wal_level=minimal\", cannot continue recovering"),
5439 : errdetail("This happens if you temporarily set \"wal_level=minimal\" on the server."),
5440 : errhint("Use a backup taken after setting \"wal_level\" to higher than \"minimal\".")));
5441 : }
5442 :
5443 : /*
5444 : * For Hot Standby, the WAL must be generated with 'replica' mode, and we
5445 : * must have at least as many backend slots as the primary.
5446 : */
5447 492 : if (ArchiveRecoveryRequested && EnableHotStandby)
5448 : {
5449 : /* We ignore autovacuum_worker_slots when we make this test. */
5450 250 : RecoveryRequiresIntParameter("max_connections",
5451 : MaxConnections,
5452 250 : ControlFile->MaxConnections);
5453 250 : RecoveryRequiresIntParameter("max_worker_processes",
5454 : max_worker_processes,
5455 250 : ControlFile->max_worker_processes);
5456 250 : RecoveryRequiresIntParameter("max_wal_senders",
5457 : max_wal_senders,
5458 250 : ControlFile->max_wal_senders);
5459 250 : RecoveryRequiresIntParameter("max_prepared_transactions",
5460 : max_prepared_xacts,
5461 250 : ControlFile->max_prepared_xacts);
5462 250 : RecoveryRequiresIntParameter("max_locks_per_transaction",
5463 : max_locks_per_xact,
5464 250 : ControlFile->max_locks_per_xact);
5465 : }
5466 492 : }
5467 :
5468 : /*
5469 : * This must be called ONCE during postmaster or standalone-backend startup
5470 : */
5471 : void
5472 1914 : StartupXLOG(void)
5473 : {
5474 : XLogCtlInsert *Insert;
5475 : CheckPoint checkPoint;
5476 : bool wasShutdown;
5477 : bool didCrash;
5478 : bool haveTblspcMap;
5479 : bool haveBackupLabel;
5480 : XLogRecPtr EndOfLog;
5481 : TimeLineID EndOfLogTLI;
5482 : TimeLineID newTLI;
5483 : bool performedWalRecovery;
5484 : EndOfWalRecoveryInfo *endOfRecoveryInfo;
5485 : XLogRecPtr abortedRecPtr;
5486 : XLogRecPtr missingContrecPtr;
5487 : TransactionId oldestActiveXID;
5488 1914 : bool promoted = false;
5489 : char timebuf[128];
5490 :
5491 : /*
5492 : * We should have an aux process resource owner to use, and we should not
5493 : * be in a transaction that's installed some other resowner.
5494 : */
5495 : Assert(AuxProcessResourceOwner != NULL);
5496 : Assert(CurrentResourceOwner == NULL ||
5497 : CurrentResourceOwner == AuxProcessResourceOwner);
5498 1914 : CurrentResourceOwner = AuxProcessResourceOwner;
5499 :
5500 : /*
5501 : * Check that contents look valid.
5502 : */
5503 1914 : if (!XRecOffIsValid(ControlFile->checkPoint))
5504 0 : ereport(FATAL,
5505 : (errcode(ERRCODE_DATA_CORRUPTED),
5506 : errmsg("control file contains invalid checkpoint location")));
5507 :
5508 1914 : switch (ControlFile->state)
5509 : {
5510 1496 : case DB_SHUTDOWNED:
5511 :
5512 : /*
5513 : * This is the expected case, so don't be chatty in standalone
5514 : * mode
5515 : */
5516 1496 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
5517 : (errmsg("database system was shut down at %s",
5518 : str_time(ControlFile->time,
5519 : timebuf, sizeof(timebuf)))));
5520 1496 : break;
5521 :
5522 64 : case DB_SHUTDOWNED_IN_RECOVERY:
5523 64 : ereport(LOG,
5524 : (errmsg("database system was shut down in recovery at %s",
5525 : str_time(ControlFile->time,
5526 : timebuf, sizeof(timebuf)))));
5527 64 : break;
5528 :
5529 0 : case DB_SHUTDOWNING:
5530 0 : ereport(LOG,
5531 : (errmsg("database system shutdown was interrupted; last known up at %s",
5532 : str_time(ControlFile->time,
5533 : timebuf, sizeof(timebuf)))));
5534 0 : break;
5535 :
5536 0 : case DB_IN_CRASH_RECOVERY:
5537 0 : ereport(LOG,
5538 : (errmsg("database system was interrupted while in recovery at %s",
5539 : str_time(ControlFile->time,
5540 : timebuf, sizeof(timebuf))),
5541 : errhint("This probably means that some data is corrupted and"
5542 : " you will have to use the last backup for recovery.")));
5543 0 : break;
5544 :
5545 12 : case DB_IN_ARCHIVE_RECOVERY:
5546 12 : ereport(LOG,
5547 : (errmsg("database system was interrupted while in recovery at log time %s",
5548 : str_time(ControlFile->checkPointCopy.time,
5549 : timebuf, sizeof(timebuf))),
5550 : errhint("If this has occurred more than once some data might be corrupted"
5551 : " and you might need to choose an earlier recovery target.")));
5552 12 : break;
5553 :
5554 342 : case DB_IN_PRODUCTION:
5555 342 : ereport(LOG,
5556 : (errmsg("database system was interrupted; last known up at %s",
5557 : str_time(ControlFile->time,
5558 : timebuf, sizeof(timebuf)))));
5559 342 : break;
5560 :
5561 0 : default:
5562 0 : ereport(FATAL,
5563 : (errcode(ERRCODE_DATA_CORRUPTED),
5564 : errmsg("control file contains invalid database cluster state")));
5565 : }
5566 :
5567 : /* This is just to allow attaching to startup process with a debugger */
5568 : #ifdef XLOG_REPLAY_DELAY
5569 : if (ControlFile->state != DB_SHUTDOWNED)
5570 : pg_usleep(60000000L);
5571 : #endif
5572 :
5573 : /*
5574 : * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5575 : * In cases where someone has performed a copy for PITR, these directories
5576 : * may have been excluded and need to be re-created.
5577 : */
5578 1914 : ValidateXLOGDirectoryStructure();
5579 :
5580 : /* Set up timeout handler needed to report startup progress. */
5581 1914 : if (!IsBootstrapProcessingMode())
5582 1814 : RegisterTimeout(STARTUP_PROGRESS_TIMEOUT,
5583 : startup_progress_timeout_handler);
5584 :
5585 : /*----------
5586 : * If we previously crashed, perform a couple of actions:
5587 : *
5588 : * - The pg_wal directory may still include some temporary WAL segments
5589 : * used when creating a new segment, so perform some clean up to not
5590 : * bloat this path. This is done first as there is no point to sync
5591 : * this temporary data.
5592 : *
5593 : * - There might be data which we had written, intending to fsync it, but
5594 : * which we had not actually fsync'd yet. Therefore, a power failure in
5595 : * the near future might cause earlier unflushed writes to be lost, even
5596 : * though more recent data written to disk from here on would be
5597 : * persisted. To avoid that, fsync the entire data directory.
5598 : */
5599 1914 : if (ControlFile->state != DB_SHUTDOWNED &&
5600 418 : ControlFile->state != DB_SHUTDOWNED_IN_RECOVERY)
5601 : {
5602 354 : RemoveTempXlogFiles();
5603 354 : SyncDataDirectory();
5604 354 : didCrash = true;
5605 : }
5606 : else
5607 1560 : didCrash = false;
5608 :
5609 : /*
5610 : * Prepare for WAL recovery if needed.
5611 : *
5612 : * InitWalRecovery analyzes the control file and the backup label file, if
5613 : * any. It updates the in-memory ControlFile buffer according to the
5614 : * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5615 : * It also applies the tablespace map file, if any.
5616 : */
5617 1914 : InitWalRecovery(ControlFile, &wasShutdown,
5618 : &haveBackupLabel, &haveTblspcMap);
5619 1914 : checkPoint = ControlFile->checkPointCopy;
5620 :
5621 : /* initialize shared memory variables from the checkpoint record */
5622 1914 : TransamVariables->nextXid = checkPoint.nextXid;
5623 1914 : TransamVariables->nextOid = checkPoint.nextOid;
5624 1914 : TransamVariables->oidCount = 0;
5625 1914 : MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5626 1914 : AdvanceOldestClogXid(checkPoint.oldestXid);
5627 1914 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5628 1914 : SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5629 1914 : SetCommitTsLimit(checkPoint.oldestCommitTsXid,
5630 : checkPoint.newestCommitTsXid);
5631 :
5632 : /*
5633 : * Clear out any old relcache cache files. This is *necessary* if we do
5634 : * any WAL replay, since that would probably result in the cache files
5635 : * being out of sync with database reality. In theory we could leave them
5636 : * in place if the database had been cleanly shut down, but it seems
5637 : * safest to just remove them always and let them be rebuilt during the
5638 : * first backend startup. These files needs to be removed from all
5639 : * directories including pg_tblspc, however the symlinks are created only
5640 : * after reading tablespace_map file in case of archive recovery from
5641 : * backup, so needs to clear old relcache files here after creating
5642 : * symlinks.
5643 : */
5644 1914 : RelationCacheInitFileRemove();
5645 :
5646 : /*
5647 : * Initialize replication slots, before there's a chance to remove
5648 : * required resources.
5649 : */
5650 1914 : StartupReplicationSlots();
5651 :
5652 : /*
5653 : * Startup logical state, needs to be setup now so we have proper data
5654 : * during crash recovery.
5655 : */
5656 1912 : StartupReorderBuffer();
5657 :
5658 : /*
5659 : * Startup CLOG. This must be done after TransamVariables->nextXid has
5660 : * been initialized and before we accept connections or begin WAL replay.
5661 : */
5662 1912 : StartupCLOG();
5663 :
5664 : /*
5665 : * Startup MultiXact. We need to do this early to be able to replay
5666 : * truncations.
5667 : */
5668 1912 : StartupMultiXact();
5669 :
5670 : /*
5671 : * Ditto for commit timestamps. Activate the facility if the setting is
5672 : * enabled in the control file, as there should be no tracking of commit
5673 : * timestamps done when the setting was disabled. This facility can be
5674 : * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5675 : */
5676 1912 : if (ControlFile->track_commit_timestamp)
5677 28 : StartupCommitTs();
5678 :
5679 : /*
5680 : * Recover knowledge about replay progress of known replication partners.
5681 : */
5682 1912 : StartupReplicationOrigin();
5683 :
5684 : /*
5685 : * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5686 : * control file. On recovery, all unlogged relations are blown away, so
5687 : * the unlogged LSN counter can be reset too.
5688 : */
5689 1912 : if (ControlFile->state == DB_SHUTDOWNED)
5690 1482 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5691 1482 : ControlFile->unloggedLSN);
5692 : else
5693 430 : pg_atomic_write_membarrier_u64(&XLogCtl->unloggedLSN,
5694 : FirstNormalUnloggedLSN);
5695 :
5696 : /*
5697 : * Copy any missing timeline history files between 'now' and the recovery
5698 : * target timeline from archive to pg_wal. While we don't need those files
5699 : * ourselves - the history file of the recovery target timeline covers all
5700 : * the previous timelines in the history too - a cascading standby server
5701 : * might be interested in them. Or, if you archive the WAL from this
5702 : * server to a different archive than the primary, it'd be good for all
5703 : * the history files to get archived there after failover, so that you can
5704 : * use one of the old timelines as a PITR target. Timeline history files
5705 : * are small, so it's better to copy them unnecessarily than not copy them
5706 : * and regret later.
5707 : */
5708 1912 : restoreTimeLineHistoryFiles(checkPoint.ThisTimeLineID, recoveryTargetTLI);
5709 :
5710 : /*
5711 : * Before running in recovery, scan pg_twophase and fill in its status to
5712 : * be able to work on entries generated by redo. Doing a scan before
5713 : * taking any recovery action has the merit to discard any 2PC files that
5714 : * are newer than the first record to replay, saving from any conflicts at
5715 : * replay. This avoids as well any subsequent scans when doing recovery
5716 : * of the on-disk two-phase data.
5717 : */
5718 1912 : restoreTwoPhaseData();
5719 :
5720 : /*
5721 : * When starting with crash recovery, reset pgstat data - it might not be
5722 : * valid. Otherwise restore pgstat data. It's safe to do this here,
5723 : * because postmaster will not yet have started any other processes.
5724 : *
5725 : * NB: Restoring replication slot stats relies on slot state to have
5726 : * already been restored from disk.
5727 : *
5728 : * TODO: With a bit of extra work we could just start with a pgstat file
5729 : * associated with the checkpoint redo location we're starting from.
5730 : */
5731 1912 : if (didCrash)
5732 354 : pgstat_discard_stats();
5733 : else
5734 1558 : pgstat_restore_stats();
5735 :
5736 1912 : lastFullPageWrites = checkPoint.fullPageWrites;
5737 :
5738 1912 : RedoRecPtr = XLogCtl->RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
5739 1912 : doPageWrites = lastFullPageWrites;
5740 :
5741 : /* REDO */
5742 1912 : if (InRecovery)
5743 : {
5744 : /* Initialize state for RecoveryInProgress() */
5745 430 : SpinLockAcquire(&XLogCtl->info_lck);
5746 430 : if (InArchiveRecovery)
5747 226 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
5748 : else
5749 204 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_CRASH;
5750 430 : SpinLockRelease(&XLogCtl->info_lck);
5751 :
5752 : /*
5753 : * Update pg_control to show that we are recovering and to show the
5754 : * selected checkpoint as the place we are starting from. We also mark
5755 : * pg_control with any minimum recovery stop point obtained from a
5756 : * backup history file.
5757 : *
5758 : * No need to hold ControlFileLock yet, we aren't up far enough.
5759 : */
5760 430 : UpdateControlFile();
5761 :
5762 : /*
5763 : * If there was a backup label file, it's done its job and the info
5764 : * has now been propagated into pg_control. We must get rid of the
5765 : * label file so that if we crash during recovery, we'll pick up at
5766 : * the latest recovery restartpoint instead of going all the way back
5767 : * to the backup start point. It seems prudent though to just rename
5768 : * the file out of the way rather than delete it completely.
5769 : */
5770 430 : if (haveBackupLabel)
5771 : {
5772 146 : unlink(BACKUP_LABEL_OLD);
5773 146 : durable_rename(BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, FATAL);
5774 : }
5775 :
5776 : /*
5777 : * If there was a tablespace_map file, it's done its job and the
5778 : * symlinks have been created. We must get rid of the map file so
5779 : * that if we crash during recovery, we don't create symlinks again.
5780 : * It seems prudent though to just rename the file out of the way
5781 : * rather than delete it completely.
5782 : */
5783 430 : if (haveTblspcMap)
5784 : {
5785 4 : unlink(TABLESPACE_MAP_OLD);
5786 4 : durable_rename(TABLESPACE_MAP, TABLESPACE_MAP_OLD, FATAL);
5787 : }
5788 :
5789 : /*
5790 : * Initialize our local copy of minRecoveryPoint. When doing crash
5791 : * recovery we want to replay up to the end of WAL. Particularly, in
5792 : * the case of a promoted standby minRecoveryPoint value in the
5793 : * control file is only updated after the first checkpoint. However,
5794 : * if the instance crashes before the first post-recovery checkpoint
5795 : * is completed then recovery will use a stale location causing the
5796 : * startup process to think that there are still invalid page
5797 : * references when checking for data consistency.
5798 : */
5799 430 : if (InArchiveRecovery)
5800 : {
5801 226 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
5802 226 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
5803 : }
5804 : else
5805 : {
5806 204 : LocalMinRecoveryPoint = InvalidXLogRecPtr;
5807 204 : LocalMinRecoveryPointTLI = 0;
5808 : }
5809 :
5810 : /* Check that the GUCs used to generate the WAL allow recovery */
5811 430 : CheckRequiredParameterValues();
5812 :
5813 : /*
5814 : * We're in recovery, so unlogged relations may be trashed and must be
5815 : * reset. This should be done BEFORE allowing Hot Standby
5816 : * connections, so that read-only backends don't try to read whatever
5817 : * garbage is left over from before.
5818 : */
5819 430 : ResetUnloggedRelations(UNLOGGED_RELATION_CLEANUP);
5820 :
5821 : /*
5822 : * Likewise, delete any saved transaction snapshot files that got left
5823 : * behind by crashed backends.
5824 : */
5825 430 : DeleteAllExportedSnapshotFiles();
5826 :
5827 : /*
5828 : * Initialize for Hot Standby, if enabled. We won't let backends in
5829 : * yet, not until we've reached the min recovery point specified in
5830 : * control file and we've established a recovery snapshot from a
5831 : * running-xacts WAL record.
5832 : */
5833 430 : if (ArchiveRecoveryRequested && EnableHotStandby)
5834 : {
5835 : TransactionId *xids;
5836 : int nxids;
5837 :
5838 214 : ereport(DEBUG1,
5839 : (errmsg_internal("initializing for hot standby")));
5840 :
5841 214 : InitRecoveryTransactionEnvironment();
5842 :
5843 214 : if (wasShutdown)
5844 52 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5845 : else
5846 162 : oldestActiveXID = checkPoint.oldestActiveXid;
5847 : Assert(TransactionIdIsValid(oldestActiveXID));
5848 :
5849 : /* Tell procarray about the range of xids it has to deal with */
5850 214 : ProcArrayInitRecovery(XidFromFullTransactionId(TransamVariables->nextXid));
5851 :
5852 : /*
5853 : * Startup subtrans only. CLOG, MultiXact and commit timestamp
5854 : * have already been started up and other SLRUs are not maintained
5855 : * during recovery and need not be started yet.
5856 : */
5857 214 : StartupSUBTRANS(oldestActiveXID);
5858 :
5859 : /*
5860 : * If we're beginning at a shutdown checkpoint, we know that
5861 : * nothing was running on the primary at this point. So fake-up an
5862 : * empty running-xacts record and use that here and now. Recover
5863 : * additional standby state for prepared transactions.
5864 : */
5865 214 : if (wasShutdown)
5866 : {
5867 : RunningTransactionsData running;
5868 : TransactionId latestCompletedXid;
5869 :
5870 : /* Update pg_subtrans entries for any prepared transactions */
5871 52 : StandbyRecoverPreparedTransactions();
5872 :
5873 : /*
5874 : * Construct a RunningTransactions snapshot representing a
5875 : * shut down server, with only prepared transactions still
5876 : * alive. We're never overflowed at this point because all
5877 : * subxids are listed with their parent prepared transactions.
5878 : */
5879 52 : running.xcnt = nxids;
5880 52 : running.subxcnt = 0;
5881 52 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
5882 52 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5883 52 : running.oldestRunningXid = oldestActiveXID;
5884 52 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5885 52 : TransactionIdRetreat(latestCompletedXid);
5886 : Assert(TransactionIdIsNormal(latestCompletedXid));
5887 52 : running.latestCompletedXid = latestCompletedXid;
5888 52 : running.xids = xids;
5889 :
5890 52 : ProcArrayApplyRecoveryInfo(&running);
5891 : }
5892 : }
5893 :
5894 : /*
5895 : * We're all set for replaying the WAL now. Do it.
5896 : */
5897 430 : PerformWalRecovery();
5898 314 : performedWalRecovery = true;
5899 : }
5900 : else
5901 1482 : performedWalRecovery = false;
5902 :
5903 : /*
5904 : * Finish WAL recovery.
5905 : */
5906 1796 : endOfRecoveryInfo = FinishWalRecovery();
5907 1796 : EndOfLog = endOfRecoveryInfo->endOfLog;
5908 1796 : EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5909 1796 : abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5910 1796 : missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5911 :
5912 : /*
5913 : * Reset ps status display, so as no information related to recovery shows
5914 : * up.
5915 : */
5916 1796 : set_ps_display("");
5917 :
5918 : /*
5919 : * When recovering from a backup (we are in recovery, and archive recovery
5920 : * was requested), complain if we did not roll forward far enough to reach
5921 : * the point where the database is consistent. For regular online
5922 : * backup-from-primary, that means reaching the end-of-backup WAL record
5923 : * (at which point we reset backupStartPoint to be Invalid), for
5924 : * backup-from-replica (which can't inject records into the WAL stream),
5925 : * that point is when we reach the minRecoveryPoint in pg_control (which
5926 : * we purposefully copy last when backing up from a replica). For
5927 : * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5928 : * or snapshot-style backups (which don't), backupEndRequired will be set
5929 : * to false.
5930 : *
5931 : * Note: it is indeed okay to look at the local variable
5932 : * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5933 : * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5934 : * been advanced beyond the WAL we processed.
5935 : */
5936 1796 : if (InRecovery &&
5937 314 : (EndOfLog < LocalMinRecoveryPoint ||
5938 314 : XLogRecPtrIsValid(ControlFile->backupStartPoint)))
5939 : {
5940 : /*
5941 : * Ran off end of WAL before reaching end-of-backup WAL record, or
5942 : * minRecoveryPoint. That's a bad sign, indicating that you tried to
5943 : * recover from an online backup but never called pg_backup_stop(), or
5944 : * you didn't archive all the WAL needed.
5945 : */
5946 0 : if (ArchiveRecoveryRequested || ControlFile->backupEndRequired)
5947 : {
5948 0 : if (XLogRecPtrIsValid(ControlFile->backupStartPoint) || ControlFile->backupEndRequired)
5949 0 : ereport(FATAL,
5950 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5951 : errmsg("WAL ends before end of online backup"),
5952 : errhint("All WAL generated while online backup was taken must be available at recovery.")));
5953 : else
5954 0 : ereport(FATAL,
5955 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5956 : errmsg("WAL ends before consistent recovery point")));
5957 : }
5958 : }
5959 :
5960 : /*
5961 : * Reset unlogged relations to the contents of their INIT fork. This is
5962 : * done AFTER recovery is complete so as to include any unlogged relations
5963 : * created during recovery, but BEFORE recovery is marked as having
5964 : * completed successfully. Otherwise we'd not retry if any of the post
5965 : * end-of-recovery steps fail.
5966 : */
5967 1796 : if (InRecovery)
5968 314 : ResetUnloggedRelations(UNLOGGED_RELATION_INIT);
5969 :
5970 : /*
5971 : * Pre-scan prepared transactions to find out the range of XIDs present.
5972 : * This information is not quite needed yet, but it is positioned here so
5973 : * as potential problems are detected before any on-disk change is done.
5974 : */
5975 1796 : oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5976 :
5977 : /*
5978 : * Allow ordinary WAL segment creation before possibly switching to a new
5979 : * timeline, which creates a new segment, and after the last ReadRecord().
5980 : */
5981 1796 : SetInstallXLogFileSegmentActive();
5982 :
5983 : /*
5984 : * Consider whether we need to assign a new timeline ID.
5985 : *
5986 : * If we did archive recovery, we always assign a new ID. This handles a
5987 : * couple of issues. If we stopped short of the end of WAL during
5988 : * recovery, then we are clearly generating a new timeline and must assign
5989 : * it a unique new ID. Even if we ran to the end, modifying the current
5990 : * last segment is problematic because it may result in trying to
5991 : * overwrite an already-archived copy of that segment, and we encourage
5992 : * DBAs to make their archive_commands reject that. We can dodge the
5993 : * problem by making the new active segment have a new timeline ID.
5994 : *
5995 : * In a normal crash recovery, we can just extend the timeline we were in.
5996 : */
5997 1796 : newTLI = endOfRecoveryInfo->lastRecTLI;
5998 1796 : if (ArchiveRecoveryRequested)
5999 : {
6000 102 : newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
6001 102 : ereport(LOG,
6002 : (errmsg("selected new timeline ID: %u", newTLI)));
6003 :
6004 : /*
6005 : * Make a writable copy of the last WAL segment. (Note that we also
6006 : * have a copy of the last block of the old WAL in
6007 : * endOfRecovery->lastPage; we will use that below.)
6008 : */
6009 102 : XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
6010 :
6011 : /*
6012 : * Remove the signal files out of the way, so that we don't
6013 : * accidentally re-enter archive recovery mode in a subsequent crash.
6014 : */
6015 102 : if (endOfRecoveryInfo->standby_signal_file_found)
6016 96 : durable_unlink(STANDBY_SIGNAL_FILE, FATAL);
6017 :
6018 102 : if (endOfRecoveryInfo->recovery_signal_file_found)
6019 6 : durable_unlink(RECOVERY_SIGNAL_FILE, FATAL);
6020 :
6021 : /*
6022 : * Write the timeline history file, and have it archived. After this
6023 : * point (or rather, as soon as the file is archived), the timeline
6024 : * will appear as "taken" in the WAL archive and to any standby
6025 : * servers. If we crash before actually switching to the new
6026 : * timeline, standby servers will nevertheless think that we switched
6027 : * to the new timeline, and will try to connect to the new timeline.
6028 : * To minimize the window for that, try to do as little as possible
6029 : * between here and writing the end-of-recovery record.
6030 : */
6031 102 : writeTimeLineHistory(newTLI, recoveryTargetTLI,
6032 : EndOfLog, endOfRecoveryInfo->recoveryStopReason);
6033 :
6034 102 : ereport(LOG,
6035 : (errmsg("archive recovery complete")));
6036 : }
6037 :
6038 : /* Save the selected TimeLineID in shared memory, too */
6039 1796 : SpinLockAcquire(&XLogCtl->info_lck);
6040 1796 : XLogCtl->InsertTimeLineID = newTLI;
6041 1796 : XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
6042 1796 : SpinLockRelease(&XLogCtl->info_lck);
6043 :
6044 : /*
6045 : * Actually, if WAL ended in an incomplete record, skip the parts that
6046 : * made it through and start writing after the portion that persisted.
6047 : * (It's critical to first write an OVERWRITE_CONTRECORD message, which
6048 : * we'll do as soon as we're open for writing new WAL.)
6049 : */
6050 1796 : if (XLogRecPtrIsValid(missingContrecPtr))
6051 : {
6052 : /*
6053 : * We should only have a missingContrecPtr if we're not switching to a
6054 : * new timeline. When a timeline switch occurs, WAL is copied from the
6055 : * old timeline to the new only up to the end of the last complete
6056 : * record, so there can't be an incomplete WAL record that we need to
6057 : * disregard.
6058 : */
6059 : Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
6060 : Assert(XLogRecPtrIsValid(abortedRecPtr));
6061 22 : EndOfLog = missingContrecPtr;
6062 : }
6063 :
6064 : /*
6065 : * Prepare to write WAL starting at EndOfLog location, and init xlog
6066 : * buffer cache using the block containing the last record from the
6067 : * previous incarnation.
6068 : */
6069 1796 : Insert = &XLogCtl->Insert;
6070 1796 : Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
6071 1796 : Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
6072 :
6073 : /*
6074 : * Tricky point here: lastPage contains the *last* block that the LastRec
6075 : * record spans, not the one it starts in. The last block is indeed the
6076 : * one we want to use.
6077 : */
6078 1796 : if (EndOfLog % XLOG_BLCKSZ != 0)
6079 : {
6080 : char *page;
6081 : int len;
6082 : int firstIdx;
6083 :
6084 1736 : firstIdx = XLogRecPtrToBufIdx(EndOfLog);
6085 1736 : len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
6086 : Assert(len < XLOG_BLCKSZ);
6087 :
6088 : /* Copy the valid part of the last block, and zero the rest */
6089 1736 : page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
6090 1736 : memcpy(page, endOfRecoveryInfo->lastPage, len);
6091 1736 : memset(page + len, 0, XLOG_BLCKSZ - len);
6092 :
6093 1736 : pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
6094 1736 : XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
6095 : }
6096 : else
6097 : {
6098 : /*
6099 : * There is no partial block to copy. Just set InitializedUpTo, and
6100 : * let the first attempt to insert a log record to initialize the next
6101 : * buffer.
6102 : */
6103 60 : XLogCtl->InitializedUpTo = EndOfLog;
6104 : }
6105 :
6106 : /*
6107 : * Update local and shared status. This is OK to do without any locks
6108 : * because no other process can be reading or writing WAL yet.
6109 : */
6110 1796 : LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
6111 1796 : pg_atomic_write_u64(&XLogCtl->logInsertResult, EndOfLog);
6112 1796 : pg_atomic_write_u64(&XLogCtl->logWriteResult, EndOfLog);
6113 1796 : pg_atomic_write_u64(&XLogCtl->logFlushResult, EndOfLog);
6114 1796 : XLogCtl->LogwrtRqst.Write = EndOfLog;
6115 1796 : XLogCtl->LogwrtRqst.Flush = EndOfLog;
6116 :
6117 : /*
6118 : * Preallocate additional log files, if wanted.
6119 : */
6120 1796 : PreallocXlogFiles(EndOfLog, newTLI);
6121 :
6122 : /*
6123 : * Okay, we're officially UP.
6124 : */
6125 1796 : InRecovery = false;
6126 :
6127 : /* start the archive_timeout timer and LSN running */
6128 1796 : XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
6129 1796 : XLogCtl->lastSegSwitchLSN = EndOfLog;
6130 :
6131 : /* also initialize latestCompletedXid, to nextXid - 1 */
6132 1796 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
6133 1796 : TransamVariables->latestCompletedXid = TransamVariables->nextXid;
6134 1796 : FullTransactionIdRetreat(&TransamVariables->latestCompletedXid);
6135 1796 : LWLockRelease(ProcArrayLock);
6136 :
6137 : /*
6138 : * Start up subtrans, if not already done for hot standby. (commit
6139 : * timestamps are started below, if necessary.)
6140 : */
6141 1796 : if (standbyState == STANDBY_DISABLED)
6142 1694 : StartupSUBTRANS(oldestActiveXID);
6143 :
6144 : /*
6145 : * Perform end of recovery actions for any SLRUs that need it.
6146 : */
6147 1796 : TrimCLOG();
6148 1796 : TrimMultiXact();
6149 :
6150 : /*
6151 : * Reload shared-memory state for prepared transactions. This needs to
6152 : * happen before renaming the last partial segment of the old timeline as
6153 : * it may be possible that we have to recover some transactions from it.
6154 : */
6155 1796 : RecoverPreparedTransactions();
6156 :
6157 : /* Shut down xlogreader */
6158 1796 : ShutdownWalRecovery();
6159 :
6160 : /* Enable WAL writes for this backend only. */
6161 1796 : LocalSetXLogInsertAllowed();
6162 :
6163 : /* If necessary, write overwrite-contrecord before doing anything else */
6164 1796 : if (XLogRecPtrIsValid(abortedRecPtr))
6165 : {
6166 : Assert(XLogRecPtrIsValid(missingContrecPtr));
6167 22 : CreateOverwriteContrecordRecord(abortedRecPtr, missingContrecPtr, newTLI);
6168 : }
6169 :
6170 : /*
6171 : * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
6172 : * record before resource manager writes cleanup WAL records or checkpoint
6173 : * record is written.
6174 : */
6175 1796 : Insert->fullPageWrites = lastFullPageWrites;
6176 1796 : UpdateFullPageWrites();
6177 :
6178 : /*
6179 : * Emit checkpoint or end-of-recovery record in XLOG, if required.
6180 : */
6181 1796 : if (performedWalRecovery)
6182 314 : promoted = PerformRecoveryXLogAction();
6183 :
6184 : /*
6185 : * If any of the critical GUCs have changed, log them before we allow
6186 : * backends to write WAL.
6187 : */
6188 1796 : XLogReportParameters();
6189 :
6190 : /* If this is archive recovery, perform post-recovery cleanup actions. */
6191 1796 : if (ArchiveRecoveryRequested)
6192 102 : CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6193 :
6194 : /*
6195 : * Local WAL inserts enabled, so it's time to finish initialization of
6196 : * commit timestamp.
6197 : */
6198 1796 : CompleteCommitTsInitialization();
6199 :
6200 : /* Clean up EndOfWalRecoveryInfo data to appease Valgrind leak checking */
6201 1796 : if (endOfRecoveryInfo->lastPage)
6202 1758 : pfree(endOfRecoveryInfo->lastPage);
6203 1796 : pfree(endOfRecoveryInfo->recoveryStopReason);
6204 1796 : pfree(endOfRecoveryInfo);
6205 :
6206 : /*
6207 : * All done with end-of-recovery actions.
6208 : *
6209 : * Now allow backends to write WAL and update the control file status in
6210 : * consequence. SharedRecoveryState, that controls if backends can write
6211 : * WAL, is updated while holding ControlFileLock to prevent other backends
6212 : * to look at an inconsistent state of the control file in shared memory.
6213 : * There is still a small window during which backends can write WAL and
6214 : * the control file is still referring to a system not in DB_IN_PRODUCTION
6215 : * state while looking at the on-disk control file.
6216 : *
6217 : * Also, we use info_lck to update SharedRecoveryState to ensure that
6218 : * there are no race conditions concerning visibility of other recent
6219 : * updates to shared memory.
6220 : */
6221 1796 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6222 1796 : ControlFile->state = DB_IN_PRODUCTION;
6223 :
6224 1796 : SpinLockAcquire(&XLogCtl->info_lck);
6225 1796 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_DONE;
6226 1796 : SpinLockRelease(&XLogCtl->info_lck);
6227 :
6228 1796 : UpdateControlFile();
6229 1796 : LWLockRelease(ControlFileLock);
6230 :
6231 : /*
6232 : * Wake up all waiters for replay LSN. They need to report an error that
6233 : * recovery was ended before reaching the target LSN.
6234 : */
6235 1796 : WaitLSNWakeup(WAIT_LSN_TYPE_REPLAY, InvalidXLogRecPtr);
6236 :
6237 : /*
6238 : * Shutdown the recovery environment. This must occur after
6239 : * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6240 : * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6241 : * any session building a snapshot will not rely on KnownAssignedXids as
6242 : * RecoveryInProgress() would return false at this stage. This is
6243 : * particularly critical for prepared 2PC transactions, that would still
6244 : * need to be included in snapshots once recovery has ended.
6245 : */
6246 1796 : if (standbyState != STANDBY_DISABLED)
6247 102 : ShutdownRecoveryTransactionEnvironment();
6248 :
6249 : /*
6250 : * If there were cascading standby servers connected to us, nudge any wal
6251 : * sender processes to notice that we've been promoted.
6252 : */
6253 1796 : WalSndWakeup(true, true);
6254 :
6255 : /*
6256 : * If this was a promotion, request an (online) checkpoint now. This isn't
6257 : * required for consistency, but the last restartpoint might be far back,
6258 : * and in case of a crash, recovering from it might take a longer than is
6259 : * appropriate now that we're not in standby mode anymore.
6260 : */
6261 1796 : if (promoted)
6262 88 : RequestCheckpoint(CHECKPOINT_FORCE);
6263 1796 : }
6264 :
6265 : /*
6266 : * Callback from PerformWalRecovery(), called when we switch from crash
6267 : * recovery to archive recovery mode. Updates the control file accordingly.
6268 : */
6269 : void
6270 4 : SwitchIntoArchiveRecovery(XLogRecPtr EndRecPtr, TimeLineID replayTLI)
6271 : {
6272 : /* initialize minRecoveryPoint to this record */
6273 4 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6274 4 : ControlFile->state = DB_IN_ARCHIVE_RECOVERY;
6275 4 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6276 : {
6277 4 : ControlFile->minRecoveryPoint = EndRecPtr;
6278 4 : ControlFile->minRecoveryPointTLI = replayTLI;
6279 : }
6280 : /* update local copy */
6281 4 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
6282 4 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
6283 :
6284 : /*
6285 : * The startup process can update its local copy of minRecoveryPoint from
6286 : * this point.
6287 : */
6288 4 : updateMinRecoveryPoint = true;
6289 :
6290 4 : UpdateControlFile();
6291 :
6292 : /*
6293 : * We update SharedRecoveryState while holding the lock on ControlFileLock
6294 : * so both states are consistent in shared memory.
6295 : */
6296 4 : SpinLockAcquire(&XLogCtl->info_lck);
6297 4 : XLogCtl->SharedRecoveryState = RECOVERY_STATE_ARCHIVE;
6298 4 : SpinLockRelease(&XLogCtl->info_lck);
6299 :
6300 4 : LWLockRelease(ControlFileLock);
6301 4 : }
6302 :
6303 : /*
6304 : * Callback from PerformWalRecovery(), called when we reach the end of backup.
6305 : * Updates the control file accordingly.
6306 : */
6307 : void
6308 146 : ReachedEndOfBackup(XLogRecPtr EndRecPtr, TimeLineID tli)
6309 : {
6310 : /*
6311 : * We have reached the end of base backup, as indicated by pg_control. The
6312 : * data on disk is now consistent (unless minRecoveryPoint is further
6313 : * ahead, which can happen if we crashed during previous recovery). Reset
6314 : * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6315 : * make sure we don't allow starting up at an earlier point even if
6316 : * recovery is stopped and restarted soon after this.
6317 : */
6318 146 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6319 :
6320 146 : if (ControlFile->minRecoveryPoint < EndRecPtr)
6321 : {
6322 138 : ControlFile->minRecoveryPoint = EndRecPtr;
6323 138 : ControlFile->minRecoveryPointTLI = tli;
6324 : }
6325 :
6326 146 : ControlFile->backupStartPoint = InvalidXLogRecPtr;
6327 146 : ControlFile->backupEndPoint = InvalidXLogRecPtr;
6328 146 : ControlFile->backupEndRequired = false;
6329 146 : UpdateControlFile();
6330 :
6331 146 : LWLockRelease(ControlFileLock);
6332 146 : }
6333 :
6334 : /*
6335 : * Perform whatever XLOG actions are necessary at end of REDO.
6336 : *
6337 : * The goal here is to make sure that we'll be able to recover properly if
6338 : * we crash again. If we choose to write a checkpoint, we'll write a shutdown
6339 : * checkpoint rather than an on-line one. This is not particularly critical,
6340 : * but since we may be assigning a new TLI, using a shutdown checkpoint allows
6341 : * us to have the rule that TLI only changes in shutdown checkpoints, which
6342 : * allows some extra error checking in xlog_redo.
6343 : */
6344 : static bool
6345 314 : PerformRecoveryXLogAction(void)
6346 : {
6347 314 : bool promoted = false;
6348 :
6349 : /*
6350 : * Perform a checkpoint to update all our recovery activity to disk.
6351 : *
6352 : * Note that we write a shutdown checkpoint rather than an on-line one.
6353 : * This is not particularly critical, but since we may be assigning a new
6354 : * TLI, using a shutdown checkpoint allows us to have the rule that TLI
6355 : * only changes in shutdown checkpoints, which allows some extra error
6356 : * checking in xlog_redo.
6357 : *
6358 : * In promotion, only create a lightweight end-of-recovery record instead
6359 : * of a full checkpoint. A checkpoint is requested later, after we're
6360 : * fully out of recovery mode and already accepting queries.
6361 : */
6362 416 : if (ArchiveRecoveryRequested && IsUnderPostmaster &&
6363 102 : PromoteIsTriggered())
6364 : {
6365 88 : promoted = true;
6366 :
6367 : /*
6368 : * Insert a special WAL record to mark the end of recovery, since we
6369 : * aren't doing a checkpoint. That means that the checkpointer process
6370 : * may likely be in the middle of a time-smoothed restartpoint and
6371 : * could continue to be for minutes after this. That sounds strange,
6372 : * but the effect is roughly the same and it would be stranger to try
6373 : * to come out of the restartpoint and then checkpoint. We request a
6374 : * checkpoint later anyway, just for safety.
6375 : */
6376 88 : CreateEndOfRecoveryRecord();
6377 : }
6378 : else
6379 : {
6380 226 : RequestCheckpoint(CHECKPOINT_END_OF_RECOVERY |
6381 : CHECKPOINT_FAST |
6382 : CHECKPOINT_WAIT);
6383 : }
6384 :
6385 314 : return promoted;
6386 : }
6387 :
6388 : /*
6389 : * Is the system still in recovery?
6390 : *
6391 : * Unlike testing InRecovery, this works in any process that's connected to
6392 : * shared memory.
6393 : */
6394 : bool
6395 161088970 : RecoveryInProgress(void)
6396 : {
6397 : /*
6398 : * We check shared state each time only until we leave recovery mode. We
6399 : * can't re-enter recovery, so there's no need to keep checking after the
6400 : * shared variable has once been seen false.
6401 : */
6402 161088970 : if (!LocalRecoveryInProgress)
6403 157114272 : return false;
6404 : else
6405 : {
6406 : /*
6407 : * use volatile pointer to make sure we make a fresh read of the
6408 : * shared variable.
6409 : */
6410 3974698 : volatile XLogCtlData *xlogctl = XLogCtl;
6411 :
6412 3974698 : LocalRecoveryInProgress = (xlogctl->SharedRecoveryState != RECOVERY_STATE_DONE);
6413 :
6414 : /*
6415 : * Note: We don't need a memory barrier when we're still in recovery.
6416 : * We might exit recovery immediately after return, so the caller
6417 : * can't rely on 'true' meaning that we're still in recovery anyway.
6418 : */
6419 :
6420 3974698 : return LocalRecoveryInProgress;
6421 : }
6422 : }
6423 :
6424 : /*
6425 : * Returns current recovery state from shared memory.
6426 : *
6427 : * This returned state is kept consistent with the contents of the control
6428 : * file. See details about the possible values of RecoveryState in xlog.h.
6429 : */
6430 : RecoveryState
6431 39288 : GetRecoveryState(void)
6432 : {
6433 : RecoveryState retval;
6434 :
6435 39288 : SpinLockAcquire(&XLogCtl->info_lck);
6436 39288 : retval = XLogCtl->SharedRecoveryState;
6437 39288 : SpinLockRelease(&XLogCtl->info_lck);
6438 :
6439 39288 : return retval;
6440 : }
6441 :
6442 : /*
6443 : * Is this process allowed to insert new WAL records?
6444 : *
6445 : * Ordinarily this is essentially equivalent to !RecoveryInProgress().
6446 : * But we also have provisions for forcing the result "true" or "false"
6447 : * within specific processes regardless of the global state.
6448 : */
6449 : bool
6450 78287792 : XLogInsertAllowed(void)
6451 : {
6452 : /*
6453 : * If value is "unconditionally true" or "unconditionally false", just
6454 : * return it. This provides the normal fast path once recovery is known
6455 : * done.
6456 : */
6457 78287792 : if (LocalXLogInsertAllowed >= 0)
6458 77036848 : return (bool) LocalXLogInsertAllowed;
6459 :
6460 : /*
6461 : * Else, must check to see if we're still in recovery.
6462 : */
6463 1250944 : if (RecoveryInProgress())
6464 1232138 : return false;
6465 :
6466 : /*
6467 : * On exit from recovery, reset to "unconditionally true", since there is
6468 : * no need to keep checking.
6469 : */
6470 18806 : LocalXLogInsertAllowed = 1;
6471 18806 : return true;
6472 : }
6473 :
6474 : /*
6475 : * Make XLogInsertAllowed() return true in the current process only.
6476 : *
6477 : * Note: it is allowed to switch LocalXLogInsertAllowed back to -1 later,
6478 : * and even call LocalSetXLogInsertAllowed() again after that.
6479 : *
6480 : * Returns the previous value of LocalXLogInsertAllowed.
6481 : */
6482 : static int
6483 1854 : LocalSetXLogInsertAllowed(void)
6484 : {
6485 1854 : int oldXLogAllowed = LocalXLogInsertAllowed;
6486 :
6487 1854 : LocalXLogInsertAllowed = 1;
6488 :
6489 1854 : return oldXLogAllowed;
6490 : }
6491 :
6492 : /*
6493 : * Return the current Redo pointer from shared memory.
6494 : *
6495 : * As a side-effect, the local RedoRecPtr copy is updated.
6496 : */
6497 : XLogRecPtr
6498 591530 : GetRedoRecPtr(void)
6499 : {
6500 : XLogRecPtr ptr;
6501 :
6502 : /*
6503 : * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6504 : * grabbed a WAL insertion lock to read the authoritative value in
6505 : * Insert->RedoRecPtr, someone might update it just after we've released
6506 : * the lock.
6507 : */
6508 591530 : SpinLockAcquire(&XLogCtl->info_lck);
6509 591530 : ptr = XLogCtl->RedoRecPtr;
6510 591530 : SpinLockRelease(&XLogCtl->info_lck);
6511 :
6512 591530 : if (RedoRecPtr < ptr)
6513 3096 : RedoRecPtr = ptr;
6514 :
6515 591530 : return RedoRecPtr;
6516 : }
6517 :
6518 : /*
6519 : * Return information needed to decide whether a modified block needs a
6520 : * full-page image to be included in the WAL record.
6521 : *
6522 : * The returned values are cached copies from backend-private memory, and
6523 : * possibly out-of-date or, indeed, uninitialized, in which case they will
6524 : * be InvalidXLogRecPtr and false, respectively. XLogInsertRecord will
6525 : * re-check them against up-to-date values, while holding the WAL insert lock.
6526 : */
6527 : void
6528 29533458 : GetFullPageWriteInfo(XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
6529 : {
6530 29533458 : *RedoRecPtr_p = RedoRecPtr;
6531 29533458 : *doPageWrites_p = doPageWrites;
6532 29533458 : }
6533 :
6534 : /*
6535 : * GetInsertRecPtr -- Returns the current insert position.
6536 : *
6537 : * NOTE: The value *actually* returned is the position of the last full
6538 : * xlog page. It lags behind the real insert position by at most 1 page.
6539 : * For that, we don't need to scan through WAL insertion locks, and an
6540 : * approximation is enough for the current usage of this function.
6541 : */
6542 : XLogRecPtr
6543 15002 : GetInsertRecPtr(void)
6544 : {
6545 : XLogRecPtr recptr;
6546 :
6547 15002 : SpinLockAcquire(&XLogCtl->info_lck);
6548 15002 : recptr = XLogCtl->LogwrtRqst.Write;
6549 15002 : SpinLockRelease(&XLogCtl->info_lck);
6550 :
6551 15002 : return recptr;
6552 : }
6553 :
6554 : /*
6555 : * GetFlushRecPtr -- Returns the current flush position, ie, the last WAL
6556 : * position known to be fsync'd to disk. This should only be used on a
6557 : * system that is known not to be in recovery.
6558 : */
6559 : XLogRecPtr
6560 464164 : GetFlushRecPtr(TimeLineID *insertTLI)
6561 : {
6562 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6563 :
6564 464164 : RefreshXLogWriteResult(LogwrtResult);
6565 :
6566 : /*
6567 : * If we're writing and flushing WAL, the time line can't be changing, so
6568 : * no lock is required.
6569 : */
6570 464164 : if (insertTLI)
6571 51388 : *insertTLI = XLogCtl->InsertTimeLineID;
6572 :
6573 464164 : return LogwrtResult.Flush;
6574 : }
6575 :
6576 : /*
6577 : * GetWALInsertionTimeLine -- Returns the current timeline of a system that
6578 : * is not in recovery.
6579 : */
6580 : TimeLineID
6581 222326 : GetWALInsertionTimeLine(void)
6582 : {
6583 : Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
6584 :
6585 : /* Since the value can't be changing, no lock is required. */
6586 222326 : return XLogCtl->InsertTimeLineID;
6587 : }
6588 :
6589 : /*
6590 : * GetWALInsertionTimeLineIfSet -- If the system is not in recovery, returns
6591 : * the WAL insertion timeline; else, returns 0. Wherever possible, use
6592 : * GetWALInsertionTimeLine() instead, since it's cheaper. Note that this
6593 : * function decides recovery has ended as soon as the insert TLI is set, which
6594 : * happens before we set XLogCtl->SharedRecoveryState to RECOVERY_STATE_DONE.
6595 : */
6596 : TimeLineID
6597 0 : GetWALInsertionTimeLineIfSet(void)
6598 : {
6599 : TimeLineID insertTLI;
6600 :
6601 0 : SpinLockAcquire(&XLogCtl->info_lck);
6602 0 : insertTLI = XLogCtl->InsertTimeLineID;
6603 0 : SpinLockRelease(&XLogCtl->info_lck);
6604 :
6605 0 : return insertTLI;
6606 : }
6607 :
6608 : /*
6609 : * GetLastImportantRecPtr -- Returns the LSN of the last important record
6610 : * inserted. All records not explicitly marked as unimportant are considered
6611 : * important.
6612 : *
6613 : * The LSN is determined by computing the maximum of
6614 : * WALInsertLocks[i].lastImportantAt.
6615 : */
6616 : XLogRecPtr
6617 3166 : GetLastImportantRecPtr(void)
6618 : {
6619 3166 : XLogRecPtr res = InvalidXLogRecPtr;
6620 : int i;
6621 :
6622 28494 : for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6623 : {
6624 : XLogRecPtr last_important;
6625 :
6626 : /*
6627 : * Need to take a lock to prevent torn reads of the LSN, which are
6628 : * possible on some of the supported platforms. WAL insert locks only
6629 : * support exclusive mode, so we have to use that.
6630 : */
6631 25328 : LWLockAcquire(&WALInsertLocks[i].l.lock, LW_EXCLUSIVE);
6632 25328 : last_important = WALInsertLocks[i].l.lastImportantAt;
6633 25328 : LWLockRelease(&WALInsertLocks[i].l.lock);
6634 :
6635 25328 : if (res < last_important)
6636 5390 : res = last_important;
6637 : }
6638 :
6639 3166 : return res;
6640 : }
6641 :
6642 : /*
6643 : * Get the time and LSN of the last xlog segment switch
6644 : */
6645 : pg_time_t
6646 0 : GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
6647 : {
6648 : pg_time_t result;
6649 :
6650 : /* Need WALWriteLock, but shared lock is sufficient */
6651 0 : LWLockAcquire(WALWriteLock, LW_SHARED);
6652 0 : result = XLogCtl->lastSegSwitchTime;
6653 0 : *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
6654 0 : LWLockRelease(WALWriteLock);
6655 :
6656 0 : return result;
6657 : }
6658 :
6659 : /*
6660 : * This must be called ONCE during postmaster or standalone-backend shutdown
6661 : */
6662 : void
6663 1314 : ShutdownXLOG(int code, Datum arg)
6664 : {
6665 : /*
6666 : * We should have an aux process resource owner to use, and we should not
6667 : * be in a transaction that's installed some other resowner.
6668 : */
6669 : Assert(AuxProcessResourceOwner != NULL);
6670 : Assert(CurrentResourceOwner == NULL ||
6671 : CurrentResourceOwner == AuxProcessResourceOwner);
6672 1314 : CurrentResourceOwner = AuxProcessResourceOwner;
6673 :
6674 : /* Don't be chatty in standalone mode */
6675 1314 : ereport(IsPostmasterEnvironment ? LOG : NOTICE,
6676 : (errmsg("shutting down")));
6677 :
6678 : /*
6679 : * Signal walsenders to move to stopping state.
6680 : */
6681 1314 : WalSndInitStopping();
6682 :
6683 : /*
6684 : * Wait for WAL senders to be in stopping state. This prevents commands
6685 : * from writing new WAL.
6686 : */
6687 1314 : WalSndWaitStopping();
6688 :
6689 1314 : if (RecoveryInProgress())
6690 110 : CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6691 : else
6692 : {
6693 : /*
6694 : * If archiving is enabled, rotate the last XLOG file so that all the
6695 : * remaining records are archived (postmaster wakes up the archiver
6696 : * process one more time at the end of shutdown). The checkpoint
6697 : * record will go to the next XLOG file and won't be archived (yet).
6698 : */
6699 1204 : if (XLogArchivingActive())
6700 28 : RequestXLogSwitch(false);
6701 :
6702 1204 : CreateCheckPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_FAST);
6703 : }
6704 1314 : }
6705 :
6706 : /*
6707 : * Log start of a checkpoint.
6708 : */
6709 : static void
6710 2874 : LogCheckpointStart(int flags, bool restartpoint)
6711 : {
6712 2874 : if (restartpoint)
6713 380 : ereport(LOG,
6714 : /* translator: the placeholders show checkpoint options */
6715 : (errmsg("restartpoint starting:%s%s%s%s%s%s%s%s",
6716 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6717 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6718 : (flags & CHECKPOINT_FAST) ? " fast" : "",
6719 : (flags & CHECKPOINT_FORCE) ? " force" : "",
6720 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6721 : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6722 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6723 : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6724 : else
6725 2494 : ereport(LOG,
6726 : /* translator: the placeholders show checkpoint options */
6727 : (errmsg("checkpoint starting:%s%s%s%s%s%s%s%s",
6728 : (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6729 : (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6730 : (flags & CHECKPOINT_FAST) ? " fast" : "",
6731 : (flags & CHECKPOINT_FORCE) ? " force" : "",
6732 : (flags & CHECKPOINT_WAIT) ? " wait" : "",
6733 : (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6734 : (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6735 : (flags & CHECKPOINT_FLUSH_UNLOGGED) ? " flush-unlogged" : "")));
6736 2874 : }
6737 :
6738 : /*
6739 : * Log end of a checkpoint.
6740 : */
6741 : static void
6742 3458 : LogCheckpointEnd(bool restartpoint)
6743 : {
6744 : long write_msecs,
6745 : sync_msecs,
6746 : total_msecs,
6747 : longest_msecs,
6748 : average_msecs;
6749 : uint64 average_sync_time;
6750 :
6751 3458 : CheckpointStats.ckpt_end_t = GetCurrentTimestamp();
6752 :
6753 3458 : write_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_write_t,
6754 : CheckpointStats.ckpt_sync_t);
6755 :
6756 3458 : sync_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_sync_t,
6757 : CheckpointStats.ckpt_sync_end_t);
6758 :
6759 : /* Accumulate checkpoint timing summary data, in milliseconds. */
6760 3458 : PendingCheckpointerStats.write_time += write_msecs;
6761 3458 : PendingCheckpointerStats.sync_time += sync_msecs;
6762 :
6763 : /*
6764 : * All of the published timing statistics are accounted for. Only
6765 : * continue if a log message is to be written.
6766 : */
6767 3458 : if (!log_checkpoints)
6768 584 : return;
6769 :
6770 2874 : total_msecs = TimestampDifferenceMilliseconds(CheckpointStats.ckpt_start_t,
6771 : CheckpointStats.ckpt_end_t);
6772 :
6773 : /*
6774 : * Timing values returned from CheckpointStats are in microseconds.
6775 : * Convert to milliseconds for consistent printing.
6776 : */
6777 2874 : longest_msecs = (long) ((CheckpointStats.ckpt_longest_sync + 999) / 1000);
6778 :
6779 2874 : average_sync_time = 0;
6780 2874 : if (CheckpointStats.ckpt_sync_rels > 0)
6781 0 : average_sync_time = CheckpointStats.ckpt_agg_sync_time /
6782 0 : CheckpointStats.ckpt_sync_rels;
6783 2874 : average_msecs = (long) ((average_sync_time + 999) / 1000);
6784 :
6785 : /*
6786 : * ControlFileLock is not required to see ControlFile->checkPoint and
6787 : * ->checkPointCopy here as we are the only updator of those variables at
6788 : * this moment.
6789 : */
6790 2874 : if (restartpoint)
6791 380 : ereport(LOG,
6792 : (errmsg("restartpoint complete: wrote %d buffers (%.1f%%), "
6793 : "wrote %d SLRU buffers; %d WAL file(s) added, "
6794 : "%d removed, %d recycled; write=%ld.%03d s, "
6795 : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6796 : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6797 : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6798 : CheckpointStats.ckpt_bufs_written,
6799 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6800 : CheckpointStats.ckpt_slru_written,
6801 : CheckpointStats.ckpt_segs_added,
6802 : CheckpointStats.ckpt_segs_removed,
6803 : CheckpointStats.ckpt_segs_recycled,
6804 : write_msecs / 1000, (int) (write_msecs % 1000),
6805 : sync_msecs / 1000, (int) (sync_msecs % 1000),
6806 : total_msecs / 1000, (int) (total_msecs % 1000),
6807 : CheckpointStats.ckpt_sync_rels,
6808 : longest_msecs / 1000, (int) (longest_msecs % 1000),
6809 : average_msecs / 1000, (int) (average_msecs % 1000),
6810 : (int) (PrevCheckPointDistance / 1024.0),
6811 : (int) (CheckPointDistanceEstimate / 1024.0),
6812 : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6813 : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6814 : else
6815 2494 : ereport(LOG,
6816 : (errmsg("checkpoint complete: wrote %d buffers (%.1f%%), "
6817 : "wrote %d SLRU buffers; %d WAL file(s) added, "
6818 : "%d removed, %d recycled; write=%ld.%03d s, "
6819 : "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6820 : "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6821 : "estimate=%d kB; lsn=%X/%08X, redo lsn=%X/%08X",
6822 : CheckpointStats.ckpt_bufs_written,
6823 : (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6824 : CheckpointStats.ckpt_slru_written,
6825 : CheckpointStats.ckpt_segs_added,
6826 : CheckpointStats.ckpt_segs_removed,
6827 : CheckpointStats.ckpt_segs_recycled,
6828 : write_msecs / 1000, (int) (write_msecs % 1000),
6829 : sync_msecs / 1000, (int) (sync_msecs % 1000),
6830 : total_msecs / 1000, (int) (total_msecs % 1000),
6831 : CheckpointStats.ckpt_sync_rels,
6832 : longest_msecs / 1000, (int) (longest_msecs % 1000),
6833 : average_msecs / 1000, (int) (average_msecs % 1000),
6834 : (int) (PrevCheckPointDistance / 1024.0),
6835 : (int) (CheckPointDistanceEstimate / 1024.0),
6836 : LSN_FORMAT_ARGS(ControlFile->checkPoint),
6837 : LSN_FORMAT_ARGS(ControlFile->checkPointCopy.redo))));
6838 : }
6839 :
6840 : /*
6841 : * Update the estimate of distance between checkpoints.
6842 : *
6843 : * The estimate is used to calculate the number of WAL segments to keep
6844 : * preallocated, see XLOGfileslop().
6845 : */
6846 : static void
6847 3458 : UpdateCheckPointDistanceEstimate(uint64 nbytes)
6848 : {
6849 : /*
6850 : * To estimate the number of segments consumed between checkpoints, keep a
6851 : * moving average of the amount of WAL generated in previous checkpoint
6852 : * cycles. However, if the load is bursty, with quiet periods and busy
6853 : * periods, we want to cater for the peak load. So instead of a plain
6854 : * moving average, let the average decline slowly if the previous cycle
6855 : * used less WAL than estimated, but bump it up immediately if it used
6856 : * more.
6857 : *
6858 : * When checkpoints are triggered by max_wal_size, this should converge to
6859 : * CheckpointSegments * wal_segment_size,
6860 : *
6861 : * Note: This doesn't pay any attention to what caused the checkpoint.
6862 : * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
6863 : * starting a base backup, are counted the same as those created
6864 : * automatically. The slow-decline will largely mask them out, if they are
6865 : * not frequent. If they are frequent, it seems reasonable to count them
6866 : * in as any others; if you issue a manual checkpoint every 5 minutes and
6867 : * never let a timed checkpoint happen, it makes sense to base the
6868 : * preallocation on that 5 minute interval rather than whatever
6869 : * checkpoint_timeout is set to.
6870 : */
6871 3458 : PrevCheckPointDistance = nbytes;
6872 3458 : if (CheckPointDistanceEstimate < nbytes)
6873 1470 : CheckPointDistanceEstimate = nbytes;
6874 : else
6875 1988 : CheckPointDistanceEstimate =
6876 1988 : (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
6877 3458 : }
6878 :
6879 : /*
6880 : * Update the ps display for a process running a checkpoint. Note that
6881 : * this routine should not do any allocations so as it can be called
6882 : * from a critical section.
6883 : */
6884 : static void
6885 6916 : update_checkpoint_display(int flags, bool restartpoint, bool reset)
6886 : {
6887 : /*
6888 : * The status is reported only for end-of-recovery and shutdown
6889 : * checkpoints or shutdown restartpoints. Updating the ps display is
6890 : * useful in those situations as it may not be possible to rely on
6891 : * pg_stat_activity to see the status of the checkpointer or the startup
6892 : * process.
6893 : */
6894 6916 : if ((flags & (CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IS_SHUTDOWN)) == 0)
6895 4304 : return;
6896 :
6897 2612 : if (reset)
6898 1306 : set_ps_display("");
6899 : else
6900 : {
6901 : char activitymsg[128];
6902 :
6903 3918 : snprintf(activitymsg, sizeof(activitymsg), "performing %s%s%s",
6904 1306 : (flags & CHECKPOINT_END_OF_RECOVERY) ? "end-of-recovery " : "",
6905 1306 : (flags & CHECKPOINT_IS_SHUTDOWN) ? "shutdown " : "",
6906 : restartpoint ? "restartpoint" : "checkpoint");
6907 1306 : set_ps_display(activitymsg);
6908 : }
6909 : }
6910 :
6911 :
6912 : /*
6913 : * Perform a checkpoint --- either during shutdown, or on-the-fly
6914 : *
6915 : * flags is a bitwise OR of the following:
6916 : * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
6917 : * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
6918 : * CHECKPOINT_FAST: finish the checkpoint ASAP, ignoring
6919 : * checkpoint_completion_target parameter.
6920 : * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
6921 : * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
6922 : * CHECKPOINT_END_OF_RECOVERY).
6923 : * CHECKPOINT_FLUSH_UNLOGGED: also flush buffers of unlogged tables.
6924 : *
6925 : * Note: flags contains other bits, of interest here only for logging purposes.
6926 : * In particular note that this routine is synchronous and does not pay
6927 : * attention to CHECKPOINT_WAIT.
6928 : *
6929 : * If !shutdown then we are writing an online checkpoint. An XLOG_CHECKPOINT_REDO
6930 : * record is inserted into WAL at the logical location of the checkpoint, before
6931 : * flushing anything to disk, and when the checkpoint is eventually completed,
6932 : * and it is from this point that WAL replay will begin in the case of a recovery
6933 : * from this checkpoint. Once everything is written to disk, an
6934 : * XLOG_CHECKPOINT_ONLINE record is written to complete the checkpoint, and
6935 : * points back to the earlier XLOG_CHECKPOINT_REDO record. This mechanism allows
6936 : * other write-ahead log records to be written while the checkpoint is in
6937 : * progress, but we must be very careful about order of operations. This function
6938 : * may take many minutes to execute on a busy system.
6939 : *
6940 : * On the other hand, when shutdown is true, concurrent insertion into the
6941 : * write-ahead log is impossible, so there is no need for two separate records.
6942 : * In this case, we only insert an XLOG_CHECKPOINT_SHUTDOWN record, and it's
6943 : * both the record marking the completion of the checkpoint and the location
6944 : * from which WAL replay would begin if needed.
6945 : *
6946 : * Returns true if a new checkpoint was performed, or false if it was skipped
6947 : * because the system was idle.
6948 : */
6949 : bool
6950 3078 : CreateCheckPoint(int flags)
6951 : {
6952 : bool shutdown;
6953 : CheckPoint checkPoint;
6954 : XLogRecPtr recptr;
6955 : XLogSegNo _logSegNo;
6956 3078 : XLogCtlInsert *Insert = &XLogCtl->Insert;
6957 : uint32 freespace;
6958 : XLogRecPtr PriorRedoPtr;
6959 : XLogRecPtr last_important_lsn;
6960 : VirtualTransactionId *vxids;
6961 : int nvxids;
6962 3078 : int oldXLogAllowed = 0;
6963 :
6964 : /*
6965 : * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6966 : * issued at a different time.
6967 : */
6968 3078 : if (flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY))
6969 1262 : shutdown = true;
6970 : else
6971 1816 : shutdown = false;
6972 :
6973 : /* sanity check */
6974 3078 : if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
6975 0 : elog(ERROR, "can't create a checkpoint during recovery");
6976 :
6977 : /*
6978 : * Prepare to accumulate statistics.
6979 : *
6980 : * Note: because it is possible for log_checkpoints to change while a
6981 : * checkpoint proceeds, we always accumulate stats, even if
6982 : * log_checkpoints is currently off.
6983 : */
6984 33858 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
6985 3078 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
6986 :
6987 : /*
6988 : * Let smgr prepare for checkpoint; this has to happen outside the
6989 : * critical section and before we determine the REDO pointer. Note that
6990 : * smgr must not do anything that'd have to be undone if we decide no
6991 : * checkpoint is needed.
6992 : */
6993 3078 : SyncPreCheckpoint();
6994 :
6995 : /*
6996 : * Use a critical section to force system panic if we have trouble.
6997 : */
6998 3078 : START_CRIT_SECTION();
6999 :
7000 3078 : if (shutdown)
7001 : {
7002 1262 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7003 1262 : ControlFile->state = DB_SHUTDOWNING;
7004 1262 : UpdateControlFile();
7005 1262 : LWLockRelease(ControlFileLock);
7006 : }
7007 :
7008 : /* Begin filling in the checkpoint WAL record */
7009 36936 : MemSet(&checkPoint, 0, sizeof(checkPoint));
7010 3078 : checkPoint.time = (pg_time_t) time(NULL);
7011 :
7012 : /*
7013 : * For Hot Standby, derive the oldestActiveXid before we fix the redo
7014 : * pointer. This allows us to begin accumulating changes to assemble our
7015 : * starting snapshot of locks and transactions.
7016 : */
7017 3078 : if (!shutdown && XLogStandbyInfoActive())
7018 1720 : checkPoint.oldestActiveXid = GetOldestActiveTransactionId(false, true);
7019 : else
7020 1358 : checkPoint.oldestActiveXid = InvalidTransactionId;
7021 :
7022 : /*
7023 : * Get location of last important record before acquiring insert locks (as
7024 : * GetLastImportantRecPtr() also locks WAL locks).
7025 : */
7026 3078 : last_important_lsn = GetLastImportantRecPtr();
7027 :
7028 : /*
7029 : * If this isn't a shutdown or forced checkpoint, and if there has been no
7030 : * WAL activity requiring a checkpoint, skip it. The idea here is to
7031 : * avoid inserting duplicate checkpoints when the system is idle.
7032 : */
7033 3078 : if ((flags & (CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_END_OF_RECOVERY |
7034 : CHECKPOINT_FORCE)) == 0)
7035 : {
7036 374 : if (last_important_lsn == ControlFile->checkPoint)
7037 : {
7038 0 : END_CRIT_SECTION();
7039 0 : ereport(DEBUG1,
7040 : (errmsg_internal("checkpoint skipped because system is idle")));
7041 0 : return false;
7042 : }
7043 : }
7044 :
7045 : /*
7046 : * An end-of-recovery checkpoint is created before anyone is allowed to
7047 : * write WAL. To allow us to write the checkpoint record, temporarily
7048 : * enable XLogInsertAllowed.
7049 : */
7050 3078 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7051 58 : oldXLogAllowed = LocalSetXLogInsertAllowed();
7052 :
7053 3078 : checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
7054 3078 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7055 58 : checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
7056 : else
7057 3020 : checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
7058 :
7059 : /*
7060 : * We must block concurrent insertions while examining insert state.
7061 : */
7062 3078 : WALInsertLockAcquireExclusive();
7063 :
7064 3078 : checkPoint.fullPageWrites = Insert->fullPageWrites;
7065 3078 : checkPoint.wal_level = wal_level;
7066 :
7067 3078 : if (shutdown)
7068 : {
7069 1262 : XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
7070 :
7071 : /*
7072 : * Compute new REDO record ptr = location of next XLOG record.
7073 : *
7074 : * Since this is a shutdown checkpoint, there can't be any concurrent
7075 : * WAL insertion.
7076 : */
7077 1262 : freespace = INSERT_FREESPACE(curInsert);
7078 1262 : if (freespace == 0)
7079 : {
7080 0 : if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
7081 0 : curInsert += SizeOfXLogLongPHD;
7082 : else
7083 0 : curInsert += SizeOfXLogShortPHD;
7084 : }
7085 1262 : checkPoint.redo = curInsert;
7086 :
7087 : /*
7088 : * Here we update the shared RedoRecPtr for future XLogInsert calls;
7089 : * this must be done while holding all the insertion locks.
7090 : *
7091 : * Note: if we fail to complete the checkpoint, RedoRecPtr will be
7092 : * left pointing past where it really needs to point. This is okay;
7093 : * the only consequence is that XLogInsert might back up whole buffers
7094 : * that it didn't really need to. We can't postpone advancing
7095 : * RedoRecPtr because XLogInserts that happen while we are dumping
7096 : * buffers must assume that their buffer changes are not included in
7097 : * the checkpoint.
7098 : */
7099 1262 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
7100 : }
7101 :
7102 : /*
7103 : * Now we can release the WAL insertion locks, allowing other xacts to
7104 : * proceed while we are flushing disk buffers.
7105 : */
7106 3078 : WALInsertLockRelease();
7107 :
7108 : /*
7109 : * If this is an online checkpoint, we have not yet determined the redo
7110 : * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
7111 : * record; the LSN at which it starts becomes the new redo pointer. We
7112 : * don't do this for a shutdown checkpoint, because in that case no WAL
7113 : * can be written between the redo point and the insertion of the
7114 : * checkpoint record itself, so the checkpoint record itself serves to
7115 : * mark the redo point.
7116 : */
7117 3078 : if (!shutdown)
7118 : {
7119 : /* Include WAL level in record for WAL summarizer's benefit. */
7120 1816 : XLogBeginInsert();
7121 1816 : XLogRegisterData(&wal_level, sizeof(wal_level));
7122 1816 : (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
7123 :
7124 : /*
7125 : * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
7126 : * shared memory and RedoRecPtr in backend-local memory, but we need
7127 : * to copy that into the record that will be inserted when the
7128 : * checkpoint is complete.
7129 : */
7130 1816 : checkPoint.redo = RedoRecPtr;
7131 : }
7132 :
7133 : /* Update the info_lck-protected copy of RedoRecPtr as well */
7134 3078 : SpinLockAcquire(&XLogCtl->info_lck);
7135 3078 : XLogCtl->RedoRecPtr = checkPoint.redo;
7136 3078 : SpinLockRelease(&XLogCtl->info_lck);
7137 :
7138 : /*
7139 : * If enabled, log checkpoint start. We postpone this until now so as not
7140 : * to log anything if we decided to skip the checkpoint.
7141 : */
7142 3078 : if (log_checkpoints)
7143 2494 : LogCheckpointStart(flags, false);
7144 :
7145 : /* Update the process title */
7146 3078 : update_checkpoint_display(flags, false, false);
7147 :
7148 : TRACE_POSTGRESQL_CHECKPOINT_START(flags);
7149 :
7150 : /*
7151 : * Get the other info we need for the checkpoint record.
7152 : *
7153 : * We don't need to save oldestClogXid in the checkpoint, it only matters
7154 : * for the short period in which clog is being truncated, and if we crash
7155 : * during that we'll redo the clog truncation and fix up oldestClogXid
7156 : * there.
7157 : */
7158 3078 : LWLockAcquire(XidGenLock, LW_SHARED);
7159 3078 : checkPoint.nextXid = TransamVariables->nextXid;
7160 3078 : checkPoint.oldestXid = TransamVariables->oldestXid;
7161 3078 : checkPoint.oldestXidDB = TransamVariables->oldestXidDB;
7162 3078 : LWLockRelease(XidGenLock);
7163 :
7164 3078 : LWLockAcquire(CommitTsLock, LW_SHARED);
7165 3078 : checkPoint.oldestCommitTsXid = TransamVariables->oldestCommitTsXid;
7166 3078 : checkPoint.newestCommitTsXid = TransamVariables->newestCommitTsXid;
7167 3078 : LWLockRelease(CommitTsLock);
7168 :
7169 3078 : LWLockAcquire(OidGenLock, LW_SHARED);
7170 3078 : checkPoint.nextOid = TransamVariables->nextOid;
7171 3078 : if (!shutdown)
7172 1816 : checkPoint.nextOid += TransamVariables->oidCount;
7173 3078 : LWLockRelease(OidGenLock);
7174 :
7175 3078 : MultiXactGetCheckptMulti(shutdown,
7176 : &checkPoint.nextMulti,
7177 : &checkPoint.nextMultiOffset,
7178 : &checkPoint.oldestMulti,
7179 : &checkPoint.oldestMultiDB);
7180 :
7181 : /*
7182 : * Having constructed the checkpoint record, ensure all shmem disk buffers
7183 : * and commit-log buffers are flushed to disk.
7184 : *
7185 : * This I/O could fail for various reasons. If so, we will fail to
7186 : * complete the checkpoint, but there is no reason to force a system
7187 : * panic. Accordingly, exit critical section while doing it.
7188 : */
7189 3078 : END_CRIT_SECTION();
7190 :
7191 : /*
7192 : * In some cases there are groups of actions that must all occur on one
7193 : * side or the other of a checkpoint record. Before flushing the
7194 : * checkpoint record we must explicitly wait for any backend currently
7195 : * performing those groups of actions.
7196 : *
7197 : * One example is end of transaction, so we must wait for any transactions
7198 : * that are currently in commit critical sections. If an xact inserted
7199 : * its commit record into XLOG just before the REDO point, then a crash
7200 : * restart from the REDO point would not replay that record, which means
7201 : * that our flushing had better include the xact's update of pg_xact. So
7202 : * we wait till he's out of his commit critical section before proceeding.
7203 : * See notes in RecordTransactionCommit().
7204 : *
7205 : * Because we've already released the insertion locks, this test is a bit
7206 : * fuzzy: it is possible that we will wait for xacts we didn't really need
7207 : * to wait for. But the delay should be short and it seems better to make
7208 : * checkpoint take a bit longer than to hold off insertions longer than
7209 : * necessary. (In fact, the whole reason we have this issue is that xact.c
7210 : * does commit record XLOG insertion and clog update as two separate steps
7211 : * protected by different locks, but again that seems best on grounds of
7212 : * minimizing lock contention.)
7213 : *
7214 : * A transaction that has not yet set delayChkptFlags when we look cannot
7215 : * be at risk, since it has not inserted its commit record yet; and one
7216 : * that's already cleared it is not at risk either, since it's done fixing
7217 : * clog and we will correctly flush the update below. So we cannot miss
7218 : * any xacts we need to wait for.
7219 : */
7220 3078 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_START);
7221 3078 : if (nvxids > 0)
7222 : {
7223 : do
7224 : {
7225 : /*
7226 : * Keep absorbing fsync requests while we wait. There could even
7227 : * be a deadlock if we don't, if the process that prevents the
7228 : * checkpoint is trying to add a request to the queue.
7229 : */
7230 46 : AbsorbSyncRequests();
7231 :
7232 46 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
7233 46 : pg_usleep(10000L); /* wait for 10 msec */
7234 46 : pgstat_report_wait_end();
7235 46 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7236 : DELAY_CHKPT_START));
7237 : }
7238 3078 : pfree(vxids);
7239 :
7240 3078 : CheckPointGuts(checkPoint.redo, flags);
7241 :
7242 3078 : vxids = GetVirtualXIDsDelayingChkpt(&nvxids, DELAY_CHKPT_COMPLETE);
7243 3078 : if (nvxids > 0)
7244 : {
7245 : do
7246 : {
7247 0 : AbsorbSyncRequests();
7248 :
7249 0 : pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
7250 0 : pg_usleep(10000L); /* wait for 10 msec */
7251 0 : pgstat_report_wait_end();
7252 0 : } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7253 : DELAY_CHKPT_COMPLETE));
7254 : }
7255 3078 : pfree(vxids);
7256 :
7257 : /*
7258 : * Take a snapshot of running transactions and write this to WAL. This
7259 : * allows us to reconstruct the state of running transactions during
7260 : * archive recovery, if required. Skip, if this info disabled.
7261 : *
7262 : * If we are shutting down, or Startup process is completing crash
7263 : * recovery we don't need to write running xact data.
7264 : */
7265 3078 : if (!shutdown && XLogStandbyInfoActive())
7266 1720 : LogStandbySnapshot();
7267 :
7268 3078 : START_CRIT_SECTION();
7269 :
7270 : /*
7271 : * Now insert the checkpoint record into XLOG.
7272 : */
7273 3078 : XLogBeginInsert();
7274 3078 : XLogRegisterData(&checkPoint, sizeof(checkPoint));
7275 3078 : recptr = XLogInsert(RM_XLOG_ID,
7276 : shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7277 : XLOG_CHECKPOINT_ONLINE);
7278 :
7279 3078 : XLogFlush(recptr);
7280 :
7281 : /*
7282 : * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7283 : * overwritten at next startup. No-one should even try, this just allows
7284 : * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7285 : * to just temporarily disable writing until the system has exited
7286 : * recovery.
7287 : */
7288 3078 : if (shutdown)
7289 : {
7290 1262 : if (flags & CHECKPOINT_END_OF_RECOVERY)
7291 58 : LocalXLogInsertAllowed = oldXLogAllowed;
7292 : else
7293 1204 : LocalXLogInsertAllowed = 0; /* never again write WAL */
7294 : }
7295 :
7296 : /*
7297 : * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7298 : * = end of actual checkpoint record.
7299 : */
7300 3078 : if (shutdown && checkPoint.redo != ProcLastRecPtr)
7301 0 : ereport(PANIC,
7302 : (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7303 :
7304 : /*
7305 : * Remember the prior checkpoint's redo ptr for
7306 : * UpdateCheckPointDistanceEstimate()
7307 : */
7308 3078 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7309 :
7310 : /*
7311 : * Update the control file.
7312 : */
7313 3078 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7314 3078 : if (shutdown)
7315 1262 : ControlFile->state = DB_SHUTDOWNED;
7316 3078 : ControlFile->checkPoint = ProcLastRecPtr;
7317 3078 : ControlFile->checkPointCopy = checkPoint;
7318 : /* crash recovery should always recover to the end of WAL */
7319 3078 : ControlFile->minRecoveryPoint = InvalidXLogRecPtr;
7320 3078 : ControlFile->minRecoveryPointTLI = 0;
7321 :
7322 : /*
7323 : * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7324 : * unused on non-shutdown checkpoints, but seems useful to store it always
7325 : * for debugging purposes.
7326 : */
7327 3078 : ControlFile->unloggedLSN = pg_atomic_read_membarrier_u64(&XLogCtl->unloggedLSN);
7328 :
7329 3078 : UpdateControlFile();
7330 3078 : LWLockRelease(ControlFileLock);
7331 :
7332 : /*
7333 : * We are now done with critical updates; no need for system panic if we
7334 : * have trouble while fooling with old log segments.
7335 : */
7336 3078 : END_CRIT_SECTION();
7337 :
7338 : /*
7339 : * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7340 : * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7341 : * where (a) we're not inside of a critical section and (b) we can be
7342 : * certain that the relevant record has been flushed to disk, which must
7343 : * happen before it can be summarized.
7344 : *
7345 : * If this is a shutdown checkpoint, then this happens reasonably
7346 : * promptly: we've only just inserted and flushed the
7347 : * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7348 : * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7349 : * record was written before we began flushing data to disk, and that
7350 : * could be many minutes ago at this point. However, we don't XLogFlush()
7351 : * after inserting that record, so we're not guaranteed that it's on disk
7352 : * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7353 : * record.
7354 : */
7355 3078 : WakeupWalSummarizer();
7356 :
7357 : /*
7358 : * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7359 : */
7360 3078 : SyncPostCheckpoint();
7361 :
7362 : /*
7363 : * Update the average distance between checkpoints if the prior checkpoint
7364 : * exists.
7365 : */
7366 3078 : if (XLogRecPtrIsValid(PriorRedoPtr))
7367 3078 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7368 :
7369 3078 : INJECTION_POINT("checkpoint-before-old-wal-removal", NULL);
7370 :
7371 : /*
7372 : * Delete old log files, those no longer needed for last checkpoint to
7373 : * prevent the disk holding the xlog from growing full.
7374 : */
7375 3078 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7376 3078 : KeepLogSeg(recptr, &_logSegNo);
7377 3078 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7378 : _logSegNo, InvalidOid,
7379 : InvalidTransactionId))
7380 : {
7381 : /*
7382 : * Some slots have been invalidated; recalculate the old-segment
7383 : * horizon, starting again from RedoRecPtr.
7384 : */
7385 6 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7386 6 : KeepLogSeg(recptr, &_logSegNo);
7387 : }
7388 3078 : _logSegNo--;
7389 3078 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7390 : checkPoint.ThisTimeLineID);
7391 :
7392 : /*
7393 : * Make more log segments if needed. (Do this after recycling old log
7394 : * segments, since that may supply some of the needed files.)
7395 : */
7396 3078 : if (!shutdown)
7397 1816 : PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7398 :
7399 : /*
7400 : * Truncate pg_subtrans if possible. We can throw away all data before
7401 : * the oldest XMIN of any running transaction. No future transaction will
7402 : * attempt to reference any pg_subtrans entry older than that (see Asserts
7403 : * in subtrans.c). During recovery, though, we mustn't do this because
7404 : * StartupSUBTRANS hasn't been called yet.
7405 : */
7406 3078 : if (!RecoveryInProgress())
7407 3020 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7408 :
7409 : /* Real work is done; log and update stats. */
7410 3078 : LogCheckpointEnd(false);
7411 :
7412 : /* Reset the process title */
7413 3078 : update_checkpoint_display(flags, false, true);
7414 :
7415 : TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7416 : NBuffers,
7417 : CheckpointStats.ckpt_segs_added,
7418 : CheckpointStats.ckpt_segs_removed,
7419 : CheckpointStats.ckpt_segs_recycled);
7420 :
7421 3078 : return true;
7422 : }
7423 :
7424 : /*
7425 : * Mark the end of recovery in WAL though without running a full checkpoint.
7426 : * We can expect that a restartpoint is likely to be in progress as we
7427 : * do this, though we are unwilling to wait for it to complete.
7428 : *
7429 : * CreateRestartPoint() allows for the case where recovery may end before
7430 : * the restartpoint completes so there is no concern of concurrent behaviour.
7431 : */
7432 : static void
7433 88 : CreateEndOfRecoveryRecord(void)
7434 : {
7435 : xl_end_of_recovery xlrec;
7436 : XLogRecPtr recptr;
7437 :
7438 : /* sanity check */
7439 88 : if (!RecoveryInProgress())
7440 0 : elog(ERROR, "can only be used to end recovery");
7441 :
7442 88 : xlrec.end_time = GetCurrentTimestamp();
7443 88 : xlrec.wal_level = wal_level;
7444 :
7445 88 : WALInsertLockAcquireExclusive();
7446 88 : xlrec.ThisTimeLineID = XLogCtl->InsertTimeLineID;
7447 88 : xlrec.PrevTimeLineID = XLogCtl->PrevTimeLineID;
7448 88 : WALInsertLockRelease();
7449 :
7450 88 : START_CRIT_SECTION();
7451 :
7452 88 : XLogBeginInsert();
7453 88 : XLogRegisterData(&xlrec, sizeof(xl_end_of_recovery));
7454 88 : recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
7455 :
7456 88 : XLogFlush(recptr);
7457 :
7458 : /*
7459 : * Update the control file so that crash recovery can follow the timeline
7460 : * changes to this point.
7461 : */
7462 88 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7463 88 : ControlFile->minRecoveryPoint = recptr;
7464 88 : ControlFile->minRecoveryPointTLI = xlrec.ThisTimeLineID;
7465 88 : UpdateControlFile();
7466 88 : LWLockRelease(ControlFileLock);
7467 :
7468 88 : END_CRIT_SECTION();
7469 88 : }
7470 :
7471 : /*
7472 : * Write an OVERWRITE_CONTRECORD message.
7473 : *
7474 : * When on WAL replay we expect a continuation record at the start of a page
7475 : * that is not there, recovery ends and WAL writing resumes at that point.
7476 : * But it's wrong to resume writing new WAL back at the start of the record
7477 : * that was broken, because downstream consumers of that WAL (physical
7478 : * replicas) are not prepared to "rewind". So the first action after
7479 : * finishing replay of all valid WAL must be to write a record of this type
7480 : * at the point where the contrecord was missing; to support xlogreader
7481 : * detecting the special case, XLP_FIRST_IS_OVERWRITE_CONTRECORD is also added
7482 : * to the page header where the record occurs. xlogreader has an ad-hoc
7483 : * mechanism to report metadata about the broken record, which is what we
7484 : * use here.
7485 : *
7486 : * At replay time, XLP_FIRST_IS_OVERWRITE_CONTRECORD instructs xlogreader to
7487 : * skip the record it was reading, and pass back the LSN of the skipped
7488 : * record, so that its caller can verify (on "replay" of that record) that the
7489 : * XLOG_OVERWRITE_CONTRECORD matches what was effectively overwritten.
7490 : *
7491 : * 'aborted_lsn' is the beginning position of the record that was incomplete.
7492 : * It is included in the WAL record. 'pagePtr' and 'newTLI' point to the
7493 : * beginning of the XLOG page where the record is to be inserted. They must
7494 : * match the current WAL insert position, they're passed here just so that we
7495 : * can verify that.
7496 : */
7497 : static XLogRecPtr
7498 22 : CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr,
7499 : TimeLineID newTLI)
7500 : {
7501 : xl_overwrite_contrecord xlrec;
7502 : XLogRecPtr recptr;
7503 : XLogPageHeader pagehdr;
7504 : XLogRecPtr startPos;
7505 :
7506 : /* sanity checks */
7507 22 : if (!RecoveryInProgress())
7508 0 : elog(ERROR, "can only be used at end of recovery");
7509 22 : if (pagePtr % XLOG_BLCKSZ != 0)
7510 0 : elog(ERROR, "invalid position for missing continuation record %X/%08X",
7511 : LSN_FORMAT_ARGS(pagePtr));
7512 :
7513 : /* The current WAL insert position should be right after the page header */
7514 22 : startPos = pagePtr;
7515 22 : if (XLogSegmentOffset(startPos, wal_segment_size) == 0)
7516 2 : startPos += SizeOfXLogLongPHD;
7517 : else
7518 20 : startPos += SizeOfXLogShortPHD;
7519 22 : recptr = GetXLogInsertRecPtr();
7520 22 : if (recptr != startPos)
7521 0 : elog(ERROR, "invalid WAL insert position %X/%08X for OVERWRITE_CONTRECORD",
7522 : LSN_FORMAT_ARGS(recptr));
7523 :
7524 22 : START_CRIT_SECTION();
7525 :
7526 : /*
7527 : * Initialize the XLOG page header (by GetXLogBuffer), and set the
7528 : * XLP_FIRST_IS_OVERWRITE_CONTRECORD flag.
7529 : *
7530 : * No other backend is allowed to write WAL yet, so acquiring the WAL
7531 : * insertion lock is just pro forma.
7532 : */
7533 22 : WALInsertLockAcquire();
7534 22 : pagehdr = (XLogPageHeader) GetXLogBuffer(pagePtr, newTLI);
7535 22 : pagehdr->xlp_info |= XLP_FIRST_IS_OVERWRITE_CONTRECORD;
7536 22 : WALInsertLockRelease();
7537 :
7538 : /*
7539 : * Insert the XLOG_OVERWRITE_CONTRECORD record as the first record on the
7540 : * page. We know it becomes the first record, because no other backend is
7541 : * allowed to write WAL yet.
7542 : */
7543 22 : XLogBeginInsert();
7544 22 : xlrec.overwritten_lsn = aborted_lsn;
7545 22 : xlrec.overwrite_time = GetCurrentTimestamp();
7546 22 : XLogRegisterData(&xlrec, sizeof(xl_overwrite_contrecord));
7547 22 : recptr = XLogInsert(RM_XLOG_ID, XLOG_OVERWRITE_CONTRECORD);
7548 :
7549 : /* check that the record was inserted to the right place */
7550 22 : if (ProcLastRecPtr != startPos)
7551 0 : elog(ERROR, "OVERWRITE_CONTRECORD was inserted to unexpected position %X/%08X",
7552 : LSN_FORMAT_ARGS(ProcLastRecPtr));
7553 :
7554 22 : XLogFlush(recptr);
7555 :
7556 22 : END_CRIT_SECTION();
7557 :
7558 22 : return recptr;
7559 : }
7560 :
7561 : /*
7562 : * Flush all data in shared memory to disk, and fsync
7563 : *
7564 : * This is the common code shared between regular checkpoints and
7565 : * recovery restartpoints.
7566 : */
7567 : static void
7568 3458 : CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
7569 : {
7570 3458 : CheckPointRelationMap();
7571 3458 : CheckPointReplicationSlots(flags & CHECKPOINT_IS_SHUTDOWN);
7572 3458 : CheckPointSnapBuild();
7573 3458 : CheckPointLogicalRewriteHeap();
7574 3458 : CheckPointReplicationOrigin();
7575 :
7576 : /* Write out all dirty data in SLRUs and the main buffer pool */
7577 : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
7578 3458 : CheckpointStats.ckpt_write_t = GetCurrentTimestamp();
7579 3458 : CheckPointCLOG();
7580 3458 : CheckPointCommitTs();
7581 3458 : CheckPointSUBTRANS();
7582 3458 : CheckPointMultiXact();
7583 3458 : CheckPointPredicate();
7584 3458 : CheckPointBuffers(flags);
7585 :
7586 : /* Perform all queued up fsyncs */
7587 : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
7588 3458 : CheckpointStats.ckpt_sync_t = GetCurrentTimestamp();
7589 3458 : ProcessSyncRequests();
7590 3458 : CheckpointStats.ckpt_sync_end_t = GetCurrentTimestamp();
7591 : TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
7592 :
7593 : /* We deliberately delay 2PC checkpointing as long as possible */
7594 3458 : CheckPointTwoPhase(checkPointRedo);
7595 3458 : }
7596 :
7597 : /*
7598 : * Save a checkpoint for recovery restart if appropriate
7599 : *
7600 : * This function is called each time a checkpoint record is read from XLOG.
7601 : * It must determine whether the checkpoint represents a safe restartpoint or
7602 : * not. If so, the checkpoint record is stashed in shared memory so that
7603 : * CreateRestartPoint can consult it. (Note that the latter function is
7604 : * executed by the checkpointer, while this one will be executed by the
7605 : * startup process.)
7606 : */
7607 : static void
7608 1426 : RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
7609 : {
7610 : /*
7611 : * Also refrain from creating a restartpoint if we have seen any
7612 : * references to non-existent pages. Restarting recovery from the
7613 : * restartpoint would not see the references, so we would lose the
7614 : * cross-check that the pages belonged to a relation that was dropped
7615 : * later.
7616 : */
7617 1426 : if (XLogHaveInvalidPages())
7618 : {
7619 0 : elog(DEBUG2,
7620 : "could not record restart point at %X/%08X because there are unresolved references to invalid pages",
7621 : LSN_FORMAT_ARGS(checkPoint->redo));
7622 0 : return;
7623 : }
7624 :
7625 : /*
7626 : * Copy the checkpoint record to shared memory, so that checkpointer can
7627 : * work out the next time it wants to perform a restartpoint.
7628 : */
7629 1426 : SpinLockAcquire(&XLogCtl->info_lck);
7630 1426 : XLogCtl->lastCheckPointRecPtr = record->ReadRecPtr;
7631 1426 : XLogCtl->lastCheckPointEndPtr = record->EndRecPtr;
7632 1426 : XLogCtl->lastCheckPoint = *checkPoint;
7633 1426 : SpinLockRelease(&XLogCtl->info_lck);
7634 : }
7635 :
7636 : /*
7637 : * Establish a restartpoint if possible.
7638 : *
7639 : * This is similar to CreateCheckPoint, but is used during WAL recovery
7640 : * to establish a point from which recovery can roll forward without
7641 : * replaying the entire recovery log.
7642 : *
7643 : * Returns true if a new restartpoint was established. We can only establish
7644 : * a restartpoint if we have replayed a safe checkpoint record since last
7645 : * restartpoint.
7646 : */
7647 : bool
7648 1182 : CreateRestartPoint(int flags)
7649 : {
7650 : XLogRecPtr lastCheckPointRecPtr;
7651 : XLogRecPtr lastCheckPointEndPtr;
7652 : CheckPoint lastCheckPoint;
7653 : XLogRecPtr PriorRedoPtr;
7654 : XLogRecPtr receivePtr;
7655 : XLogRecPtr replayPtr;
7656 : TimeLineID replayTLI;
7657 : XLogRecPtr endptr;
7658 : XLogSegNo _logSegNo;
7659 : TimestampTz xtime;
7660 :
7661 : /* Concurrent checkpoint/restartpoint cannot happen */
7662 : Assert(!IsUnderPostmaster || MyBackendType == B_CHECKPOINTER);
7663 :
7664 : /* Get a local copy of the last safe checkpoint record. */
7665 1182 : SpinLockAcquire(&XLogCtl->info_lck);
7666 1182 : lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
7667 1182 : lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
7668 1182 : lastCheckPoint = XLogCtl->lastCheckPoint;
7669 1182 : SpinLockRelease(&XLogCtl->info_lck);
7670 :
7671 : /*
7672 : * Check that we're still in recovery mode. It's ok if we exit recovery
7673 : * mode after this check, the restart point is valid anyway.
7674 : */
7675 1182 : if (!RecoveryInProgress())
7676 : {
7677 0 : ereport(DEBUG2,
7678 : (errmsg_internal("skipping restartpoint, recovery has already ended")));
7679 0 : return false;
7680 : }
7681 :
7682 : /*
7683 : * If the last checkpoint record we've replayed is already our last
7684 : * restartpoint, we can't perform a new restart point. We still update
7685 : * minRecoveryPoint in that case, so that if this is a shutdown restart
7686 : * point, we won't start up earlier than before. That's not strictly
7687 : * necessary, but when hot standby is enabled, it would be rather weird if
7688 : * the database opened up for read-only connections at a point-in-time
7689 : * before the last shutdown. Such time travel is still possible in case of
7690 : * immediate shutdown, though.
7691 : *
7692 : * We don't explicitly advance minRecoveryPoint when we do create a
7693 : * restartpoint. It's assumed that flushing the buffers will do that as a
7694 : * side-effect.
7695 : */
7696 1182 : if (!XLogRecPtrIsValid(lastCheckPointRecPtr) ||
7697 518 : lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7698 : {
7699 802 : ereport(DEBUG2,
7700 : errmsg_internal("skipping restartpoint, already performed at %X/%08X",
7701 : LSN_FORMAT_ARGS(lastCheckPoint.redo)));
7702 :
7703 802 : UpdateMinRecoveryPoint(InvalidXLogRecPtr, true);
7704 802 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7705 : {
7706 66 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7707 66 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7708 66 : UpdateControlFile();
7709 66 : LWLockRelease(ControlFileLock);
7710 : }
7711 802 : return false;
7712 : }
7713 :
7714 : /*
7715 : * Update the shared RedoRecPtr so that the startup process can calculate
7716 : * the number of segments replayed since last restartpoint, and request a
7717 : * restartpoint if it exceeds CheckPointSegments.
7718 : *
7719 : * Like in CreateCheckPoint(), hold off insertions to update it, although
7720 : * during recovery this is just pro forma, because no WAL insertions are
7721 : * happening.
7722 : */
7723 380 : WALInsertLockAcquireExclusive();
7724 380 : RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
7725 380 : WALInsertLockRelease();
7726 :
7727 : /* Also update the info_lck-protected copy */
7728 380 : SpinLockAcquire(&XLogCtl->info_lck);
7729 380 : XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7730 380 : SpinLockRelease(&XLogCtl->info_lck);
7731 :
7732 : /*
7733 : * Prepare to accumulate statistics.
7734 : *
7735 : * Note: because it is possible for log_checkpoints to change while a
7736 : * checkpoint proceeds, we always accumulate stats, even if
7737 : * log_checkpoints is currently off.
7738 : */
7739 4180 : MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
7740 380 : CheckpointStats.ckpt_start_t = GetCurrentTimestamp();
7741 :
7742 380 : if (log_checkpoints)
7743 380 : LogCheckpointStart(flags, true);
7744 :
7745 : /* Update the process title */
7746 380 : update_checkpoint_display(flags, true, false);
7747 :
7748 380 : CheckPointGuts(lastCheckPoint.redo, flags);
7749 :
7750 : /*
7751 : * This location needs to be after CheckPointGuts() to ensure that some
7752 : * work has already happened during this checkpoint.
7753 : */
7754 380 : INJECTION_POINT("create-restart-point", NULL);
7755 :
7756 : /*
7757 : * Remember the prior checkpoint's redo ptr for
7758 : * UpdateCheckPointDistanceEstimate()
7759 : */
7760 380 : PriorRedoPtr = ControlFile->checkPointCopy.redo;
7761 :
7762 : /*
7763 : * Update pg_control, using current time. Check that it still shows an
7764 : * older checkpoint, else do nothing; this is a quick hack to make sure
7765 : * nothing really bad happens if somehow we get here after the
7766 : * end-of-recovery checkpoint.
7767 : */
7768 380 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7769 380 : if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7770 : {
7771 : /*
7772 : * Update the checkpoint information. We do this even if the cluster
7773 : * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7774 : * segments recycled below.
7775 : */
7776 380 : ControlFile->checkPoint = lastCheckPointRecPtr;
7777 380 : ControlFile->checkPointCopy = lastCheckPoint;
7778 :
7779 : /*
7780 : * Ensure minRecoveryPoint is past the checkpoint record and update it
7781 : * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7782 : * this will have happened already while writing out dirty buffers,
7783 : * but not necessarily - e.g. because no buffers were dirtied. We do
7784 : * this because a backup performed in recovery uses minRecoveryPoint
7785 : * to determine which WAL files must be included in the backup, and
7786 : * the file (or files) containing the checkpoint record must be
7787 : * included, at a minimum. Note that for an ordinary restart of
7788 : * recovery there's no value in having the minimum recovery point any
7789 : * earlier than this anyway, because redo will begin just after the
7790 : * checkpoint record.
7791 : */
7792 380 : if (ControlFile->state == DB_IN_ARCHIVE_RECOVERY)
7793 : {
7794 380 : if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7795 : {
7796 38 : ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7797 38 : ControlFile->minRecoveryPointTLI = lastCheckPoint.ThisTimeLineID;
7798 :
7799 : /* update local copy */
7800 38 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
7801 38 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
7802 : }
7803 380 : if (flags & CHECKPOINT_IS_SHUTDOWN)
7804 44 : ControlFile->state = DB_SHUTDOWNED_IN_RECOVERY;
7805 : }
7806 380 : UpdateControlFile();
7807 : }
7808 380 : LWLockRelease(ControlFileLock);
7809 :
7810 : /*
7811 : * Update the average distance between checkpoints/restartpoints if the
7812 : * prior checkpoint exists.
7813 : */
7814 380 : if (XLogRecPtrIsValid(PriorRedoPtr))
7815 380 : UpdateCheckPointDistanceEstimate(RedoRecPtr - PriorRedoPtr);
7816 :
7817 : /*
7818 : * Delete old log files, those no longer needed for last restartpoint to
7819 : * prevent the disk holding the xlog from growing full.
7820 : */
7821 380 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7822 :
7823 : /*
7824 : * Retreat _logSegNo using the current end of xlog replayed or received,
7825 : * whichever is later.
7826 : */
7827 380 : receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
7828 380 : replayPtr = GetXLogReplayRecPtr(&replayTLI);
7829 380 : endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7830 380 : KeepLogSeg(endptr, &_logSegNo);
7831 380 : if (InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_REMOVED | RS_INVAL_IDLE_TIMEOUT,
7832 : _logSegNo, InvalidOid,
7833 : InvalidTransactionId))
7834 : {
7835 : /*
7836 : * Some slots have been invalidated; recalculate the old-segment
7837 : * horizon, starting again from RedoRecPtr.
7838 : */
7839 2 : XLByteToSeg(RedoRecPtr, _logSegNo, wal_segment_size);
7840 2 : KeepLogSeg(endptr, &_logSegNo);
7841 : }
7842 380 : _logSegNo--;
7843 :
7844 : /*
7845 : * Try to recycle segments on a useful timeline. If we've been promoted
7846 : * since the beginning of this restartpoint, use the new timeline chosen
7847 : * at end of recovery. If we're still in recovery, use the timeline we're
7848 : * currently replaying.
7849 : *
7850 : * There is no guarantee that the WAL segments will be useful on the
7851 : * current timeline; if recovery proceeds to a new timeline right after
7852 : * this, the pre-allocated WAL segments on this timeline will not be used,
7853 : * and will go wasted until recycled on the next restartpoint. We'll live
7854 : * with that.
7855 : */
7856 380 : if (!RecoveryInProgress())
7857 0 : replayTLI = XLogCtl->InsertTimeLineID;
7858 :
7859 380 : RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7860 :
7861 : /*
7862 : * Make more log segments if needed. (Do this after recycling old log
7863 : * segments, since that may supply some of the needed files.)
7864 : */
7865 380 : PreallocXlogFiles(endptr, replayTLI);
7866 :
7867 : /*
7868 : * Truncate pg_subtrans if possible. We can throw away all data before
7869 : * the oldest XMIN of any running transaction. No future transaction will
7870 : * attempt to reference any pg_subtrans entry older than that (see Asserts
7871 : * in subtrans.c). When hot standby is disabled, though, we mustn't do
7872 : * this because StartupSUBTRANS hasn't been called yet.
7873 : */
7874 380 : if (EnableHotStandby)
7875 380 : TruncateSUBTRANS(GetOldestTransactionIdConsideredRunning());
7876 :
7877 : /* Real work is done; log and update stats. */
7878 380 : LogCheckpointEnd(true);
7879 :
7880 : /* Reset the process title */
7881 380 : update_checkpoint_display(flags, true, true);
7882 :
7883 380 : xtime = GetLatestXTime();
7884 380 : ereport((log_checkpoints ? LOG : DEBUG2),
7885 : errmsg("recovery restart point at %X/%08X",
7886 : LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7887 : xtime ? errdetail("Last completed transaction was at log time %s.",
7888 : timestamptz_to_str(xtime)) : 0);
7889 :
7890 : /*
7891 : * Finally, execute archive_cleanup_command, if any.
7892 : */
7893 380 : if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
7894 0 : ExecuteRecoveryCommand(archiveCleanupCommand,
7895 : "archive_cleanup_command",
7896 : false,
7897 : WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7898 :
7899 380 : return true;
7900 : }
7901 :
7902 : /*
7903 : * Report availability of WAL for the given target LSN
7904 : * (typically a slot's restart_lsn)
7905 : *
7906 : * Returns one of the following enum values:
7907 : *
7908 : * * WALAVAIL_RESERVED means targetLSN is available and it is in the range of
7909 : * max_wal_size.
7910 : *
7911 : * * WALAVAIL_EXTENDED means it is still available by preserving extra
7912 : * segments beyond max_wal_size. If max_slot_wal_keep_size is smaller
7913 : * than max_wal_size, this state is not returned.
7914 : *
7915 : * * WALAVAIL_UNRESERVED means it is being lost and the next checkpoint will
7916 : * remove reserved segments. The walsender using this slot may return to the
7917 : * above.
7918 : *
7919 : * * WALAVAIL_REMOVED means it has been removed. A replication stream on
7920 : * a slot with this LSN cannot continue. (Any associated walsender
7921 : * processes should have been terminated already.)
7922 : *
7923 : * * WALAVAIL_INVALID_LSN means the slot hasn't been set to reserve WAL.
7924 : */
7925 : WALAvailability
7926 1028 : GetWALAvailability(XLogRecPtr targetLSN)
7927 : {
7928 : XLogRecPtr currpos; /* current write LSN */
7929 : XLogSegNo currSeg; /* segid of currpos */
7930 : XLogSegNo targetSeg; /* segid of targetLSN */
7931 : XLogSegNo oldestSeg; /* actual oldest segid */
7932 : XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7933 : XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7934 : uint64 keepSegs;
7935 :
7936 : /*
7937 : * slot does not reserve WAL. Either deactivated, or has never been active
7938 : */
7939 1028 : if (!XLogRecPtrIsValid(targetLSN))
7940 70 : return WALAVAIL_INVALID_LSN;
7941 :
7942 : /*
7943 : * Calculate the oldest segment currently reserved by all slots,
7944 : * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7945 : * oldestSlotSeg to the current segment.
7946 : */
7947 958 : currpos = GetXLogWriteRecPtr();
7948 958 : XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
7949 958 : KeepLogSeg(currpos, &oldestSlotSeg);
7950 :
7951 : /*
7952 : * Find the oldest extant segment file. We get 1 until checkpoint removes
7953 : * the first WAL segment file since startup, which causes the status being
7954 : * wrong under certain abnormal conditions but that doesn't actually harm.
7955 : */
7956 958 : oldestSeg = XLogGetLastRemovedSegno() + 1;
7957 :
7958 : /* calculate oldest segment by max_wal_size */
7959 958 : XLByteToSeg(currpos, currSeg, wal_segment_size);
7960 958 : keepSegs = ConvertToXSegs(max_wal_size_mb, wal_segment_size) + 1;
7961 :
7962 958 : if (currSeg > keepSegs)
7963 16 : oldestSegMaxWalSize = currSeg - keepSegs;
7964 : else
7965 942 : oldestSegMaxWalSize = 1;
7966 :
7967 : /* the segment we care about */
7968 958 : XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7969 :
7970 : /*
7971 : * No point in returning reserved or extended status values if the
7972 : * targetSeg is known to be lost.
7973 : */
7974 958 : if (targetSeg >= oldestSlotSeg)
7975 : {
7976 : /* show "reserved" when targetSeg is within max_wal_size */
7977 956 : if (targetSeg >= oldestSegMaxWalSize)
7978 952 : return WALAVAIL_RESERVED;
7979 :
7980 : /* being retained by slots exceeding max_wal_size */
7981 4 : return WALAVAIL_EXTENDED;
7982 : }
7983 :
7984 : /* WAL segments are no longer retained but haven't been removed yet */
7985 2 : if (targetSeg >= oldestSeg)
7986 2 : return WALAVAIL_UNRESERVED;
7987 :
7988 : /* Definitely lost */
7989 0 : return WALAVAIL_REMOVED;
7990 : }
7991 :
7992 :
7993 : /*
7994 : * Retreat *logSegNo to the last segment that we need to retain because of
7995 : * either wal_keep_size or replication slots.
7996 : *
7997 : * This is calculated by subtracting wal_keep_size from the given xlog
7998 : * location, recptr and by making sure that that result is below the
7999 : * requirement of replication slots. For the latter criterion we do consider
8000 : * the effects of max_slot_wal_keep_size: reserve at most that much space back
8001 : * from recptr.
8002 : *
8003 : * Note about replication slots: if this function calculates a value
8004 : * that's further ahead than what slots need reserved, then affected
8005 : * slots need to be invalidated and this function invoked again.
8006 : * XXX it might be a good idea to rewrite this function so that
8007 : * invalidation is optionally done here, instead.
8008 : */
8009 : static void
8010 4424 : KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
8011 : {
8012 : XLogSegNo currSegNo;
8013 : XLogSegNo segno;
8014 : XLogRecPtr keep;
8015 :
8016 4424 : XLByteToSeg(recptr, currSegNo, wal_segment_size);
8017 4424 : segno = currSegNo;
8018 :
8019 : /* Calculate how many segments are kept by slots. */
8020 4424 : keep = XLogGetReplicationSlotMinimumLSN();
8021 4424 : if (XLogRecPtrIsValid(keep) && keep < recptr)
8022 : {
8023 1268 : XLByteToSeg(keep, segno, wal_segment_size);
8024 :
8025 : /*
8026 : * Account for max_slot_wal_keep_size to avoid keeping more than
8027 : * configured. However, don't do that during a binary upgrade: if
8028 : * slots were to be invalidated because of this, it would not be
8029 : * possible to preserve logical ones during the upgrade.
8030 : */
8031 1268 : if (max_slot_wal_keep_size_mb >= 0 && !IsBinaryUpgrade)
8032 : {
8033 : uint64 slot_keep_segs;
8034 :
8035 40 : slot_keep_segs =
8036 40 : ConvertToXSegs(max_slot_wal_keep_size_mb, wal_segment_size);
8037 :
8038 40 : if (currSegNo - segno > slot_keep_segs)
8039 10 : segno = currSegNo - slot_keep_segs;
8040 : }
8041 : }
8042 :
8043 : /*
8044 : * If WAL summarization is in use, don't remove WAL that has yet to be
8045 : * summarized.
8046 : */
8047 4424 : keep = GetOldestUnsummarizedLSN(NULL, NULL);
8048 4424 : if (XLogRecPtrIsValid(keep))
8049 : {
8050 : XLogSegNo unsummarized_segno;
8051 :
8052 4 : XLByteToSeg(keep, unsummarized_segno, wal_segment_size);
8053 4 : if (unsummarized_segno < segno)
8054 4 : segno = unsummarized_segno;
8055 : }
8056 :
8057 : /* but, keep at least wal_keep_size if that's set */
8058 4424 : if (wal_keep_size_mb > 0)
8059 : {
8060 : uint64 keep_segs;
8061 :
8062 148 : keep_segs = ConvertToXSegs(wal_keep_size_mb, wal_segment_size);
8063 148 : if (currSegNo - segno < keep_segs)
8064 : {
8065 : /* avoid underflow, don't go below 1 */
8066 148 : if (currSegNo <= keep_segs)
8067 140 : segno = 1;
8068 : else
8069 8 : segno = currSegNo - keep_segs;
8070 : }
8071 : }
8072 :
8073 : /* don't delete WAL segments newer than the calculated segment */
8074 4424 : if (segno < *logSegNo)
8075 750 : *logSegNo = segno;
8076 4424 : }
8077 :
8078 : /*
8079 : * Write a NEXTOID log record
8080 : */
8081 : void
8082 1226 : XLogPutNextOid(Oid nextOid)
8083 : {
8084 1226 : XLogBeginInsert();
8085 1226 : XLogRegisterData(&nextOid, sizeof(Oid));
8086 1226 : (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
8087 :
8088 : /*
8089 : * We need not flush the NEXTOID record immediately, because any of the
8090 : * just-allocated OIDs could only reach disk as part of a tuple insert or
8091 : * update that would have its own XLOG record that must follow the NEXTOID
8092 : * record. Therefore, the standard buffer LSN interlock applied to those
8093 : * records will ensure no such OID reaches disk before the NEXTOID record
8094 : * does.
8095 : *
8096 : * Note, however, that the above statement only covers state "within" the
8097 : * database. When we use a generated OID as a file or directory name, we
8098 : * are in a sense violating the basic WAL rule, because that filesystem
8099 : * change may reach disk before the NEXTOID WAL record does. The impact
8100 : * of this is that if a database crash occurs immediately afterward, we
8101 : * might after restart re-generate the same OID and find that it conflicts
8102 : * with the leftover file or directory. But since for safety's sake we
8103 : * always loop until finding a nonconflicting filename, this poses no real
8104 : * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
8105 : */
8106 1226 : }
8107 :
8108 : /*
8109 : * Write an XLOG SWITCH record.
8110 : *
8111 : * Here we just blindly issue an XLogInsert request for the record.
8112 : * All the magic happens inside XLogInsert.
8113 : *
8114 : * The return value is either the end+1 address of the switch record,
8115 : * or the end+1 address of the prior segment if we did not need to
8116 : * write a switch record because we are already at segment start.
8117 : */
8118 : XLogRecPtr
8119 1566 : RequestXLogSwitch(bool mark_unimportant)
8120 : {
8121 : XLogRecPtr RecPtr;
8122 :
8123 : /* XLOG SWITCH has no data */
8124 1566 : XLogBeginInsert();
8125 :
8126 1566 : if (mark_unimportant)
8127 0 : XLogSetRecordFlags(XLOG_MARK_UNIMPORTANT);
8128 1566 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
8129 :
8130 1566 : return RecPtr;
8131 : }
8132 :
8133 : /*
8134 : * Write a RESTORE POINT record
8135 : */
8136 : XLogRecPtr
8137 6 : XLogRestorePoint(const char *rpName)
8138 : {
8139 : XLogRecPtr RecPtr;
8140 : xl_restore_point xlrec;
8141 :
8142 6 : xlrec.rp_time = GetCurrentTimestamp();
8143 6 : strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
8144 :
8145 6 : XLogBeginInsert();
8146 6 : XLogRegisterData(&xlrec, sizeof(xl_restore_point));
8147 :
8148 6 : RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
8149 :
8150 6 : ereport(LOG,
8151 : errmsg("restore point \"%s\" created at %X/%08X",
8152 : rpName, LSN_FORMAT_ARGS(RecPtr)));
8153 :
8154 6 : return RecPtr;
8155 : }
8156 :
8157 : /*
8158 : * Check if any of the GUC parameters that are critical for hot standby
8159 : * have changed, and update the value in pg_control file if necessary.
8160 : */
8161 : static void
8162 1796 : XLogReportParameters(void)
8163 : {
8164 1796 : if (wal_level != ControlFile->wal_level ||
8165 1324 : wal_log_hints != ControlFile->wal_log_hints ||
8166 1154 : MaxConnections != ControlFile->MaxConnections ||
8167 1152 : max_worker_processes != ControlFile->max_worker_processes ||
8168 1150 : max_wal_senders != ControlFile->max_wal_senders ||
8169 1108 : max_prepared_xacts != ControlFile->max_prepared_xacts ||
8170 916 : max_locks_per_xact != ControlFile->max_locks_per_xact ||
8171 916 : track_commit_timestamp != ControlFile->track_commit_timestamp)
8172 : {
8173 : /*
8174 : * The change in number of backend slots doesn't need to be WAL-logged
8175 : * if archiving is not enabled, as you can't start archive recovery
8176 : * with wal_level=minimal anyway. We don't really care about the
8177 : * values in pg_control either if wal_level=minimal, but seems better
8178 : * to keep them up-to-date to avoid confusion.
8179 : */
8180 902 : if (wal_level != ControlFile->wal_level || XLogIsNeeded())
8181 : {
8182 : xl_parameter_change xlrec;
8183 : XLogRecPtr recptr;
8184 :
8185 862 : xlrec.MaxConnections = MaxConnections;
8186 862 : xlrec.max_worker_processes = max_worker_processes;
8187 862 : xlrec.max_wal_senders = max_wal_senders;
8188 862 : xlrec.max_prepared_xacts = max_prepared_xacts;
8189 862 : xlrec.max_locks_per_xact = max_locks_per_xact;
8190 862 : xlrec.wal_level = wal_level;
8191 862 : xlrec.wal_log_hints = wal_log_hints;
8192 862 : xlrec.track_commit_timestamp = track_commit_timestamp;
8193 :
8194 862 : XLogBeginInsert();
8195 862 : XLogRegisterData(&xlrec, sizeof(xlrec));
8196 :
8197 862 : recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
8198 862 : XLogFlush(recptr);
8199 : }
8200 :
8201 902 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8202 :
8203 902 : ControlFile->MaxConnections = MaxConnections;
8204 902 : ControlFile->max_worker_processes = max_worker_processes;
8205 902 : ControlFile->max_wal_senders = max_wal_senders;
8206 902 : ControlFile->max_prepared_xacts = max_prepared_xacts;
8207 902 : ControlFile->max_locks_per_xact = max_locks_per_xact;
8208 902 : ControlFile->wal_level = wal_level;
8209 902 : ControlFile->wal_log_hints = wal_log_hints;
8210 902 : ControlFile->track_commit_timestamp = track_commit_timestamp;
8211 902 : UpdateControlFile();
8212 :
8213 902 : LWLockRelease(ControlFileLock);
8214 : }
8215 1796 : }
8216 :
8217 : /*
8218 : * Update full_page_writes in shared memory, and write an
8219 : * XLOG_FPW_CHANGE record if necessary.
8220 : *
8221 : * Note: this function assumes there is no other process running
8222 : * concurrently that could update it.
8223 : */
8224 : void
8225 2998 : UpdateFullPageWrites(void)
8226 : {
8227 2998 : XLogCtlInsert *Insert = &XLogCtl->Insert;
8228 : bool recoveryInProgress;
8229 :
8230 : /*
8231 : * Do nothing if full_page_writes has not been changed.
8232 : *
8233 : * It's safe to check the shared full_page_writes without the lock,
8234 : * because we assume that there is no concurrently running process which
8235 : * can update it.
8236 : */
8237 2998 : if (fullPageWrites == Insert->fullPageWrites)
8238 2240 : return;
8239 :
8240 : /*
8241 : * Perform this outside critical section so that the WAL insert
8242 : * initialization done by RecoveryInProgress() doesn't trigger an
8243 : * assertion failure.
8244 : */
8245 758 : recoveryInProgress = RecoveryInProgress();
8246 :
8247 758 : START_CRIT_SECTION();
8248 :
8249 : /*
8250 : * It's always safe to take full page images, even when not strictly
8251 : * required, but not the other round. So if we're setting full_page_writes
8252 : * to true, first set it true and then write the WAL record. If we're
8253 : * setting it to false, first write the WAL record and then set the global
8254 : * flag.
8255 : */
8256 758 : if (fullPageWrites)
8257 : {
8258 738 : WALInsertLockAcquireExclusive();
8259 738 : Insert->fullPageWrites = true;
8260 738 : WALInsertLockRelease();
8261 : }
8262 :
8263 : /*
8264 : * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8265 : * full_page_writes during archive recovery, if required.
8266 : */
8267 758 : if (XLogStandbyInfoActive() && !recoveryInProgress)
8268 : {
8269 0 : XLogBeginInsert();
8270 0 : XLogRegisterData(&fullPageWrites, sizeof(bool));
8271 :
8272 0 : XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8273 : }
8274 :
8275 758 : if (!fullPageWrites)
8276 : {
8277 20 : WALInsertLockAcquireExclusive();
8278 20 : Insert->fullPageWrites = false;
8279 20 : WALInsertLockRelease();
8280 : }
8281 758 : END_CRIT_SECTION();
8282 : }
8283 :
8284 : /*
8285 : * XLOG resource manager's routines
8286 : *
8287 : * Definitions of info values are in include/catalog/pg_control.h, though
8288 : * not all record types are related to control file updates.
8289 : *
8290 : * NOTE: Some XLOG record types that are directly related to WAL recovery
8291 : * are handled in xlogrecovery_redo().
8292 : */
8293 : void
8294 87576 : xlog_redo(XLogReaderState *record)
8295 : {
8296 87576 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8297 87576 : XLogRecPtr lsn = record->EndRecPtr;
8298 :
8299 : /*
8300 : * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8301 : * XLOG_FPI_FOR_HINT records.
8302 : */
8303 : Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
8304 : !XLogRecHasAnyBlockRefs(record));
8305 :
8306 87576 : if (info == XLOG_NEXTOID)
8307 : {
8308 : Oid nextOid;
8309 :
8310 : /*
8311 : * We used to try to take the maximum of TransamVariables->nextOid and
8312 : * the recorded nextOid, but that fails if the OID counter wraps
8313 : * around. Since no OID allocation should be happening during replay
8314 : * anyway, better to just believe the record exactly. We still take
8315 : * OidGenLock while setting the variable, just in case.
8316 : */
8317 186 : memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
8318 186 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8319 186 : TransamVariables->nextOid = nextOid;
8320 186 : TransamVariables->oidCount = 0;
8321 186 : LWLockRelease(OidGenLock);
8322 : }
8323 87390 : else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8324 : {
8325 : CheckPoint checkPoint;
8326 : TimeLineID replayTLI;
8327 :
8328 70 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8329 : /* In a SHUTDOWN checkpoint, believe the counters exactly */
8330 70 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8331 70 : TransamVariables->nextXid = checkPoint.nextXid;
8332 70 : LWLockRelease(XidGenLock);
8333 70 : LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8334 70 : TransamVariables->nextOid = checkPoint.nextOid;
8335 70 : TransamVariables->oidCount = 0;
8336 70 : LWLockRelease(OidGenLock);
8337 70 : MultiXactSetNextMXact(checkPoint.nextMulti,
8338 : checkPoint.nextMultiOffset);
8339 :
8340 70 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8341 : checkPoint.oldestMultiDB);
8342 :
8343 : /*
8344 : * No need to set oldestClogXid here as well; it'll be set when we
8345 : * redo an xl_clog_truncate if it changed since initialization.
8346 : */
8347 70 : SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8348 :
8349 : /*
8350 : * If we see a shutdown checkpoint while waiting for an end-of-backup
8351 : * record, the backup was canceled and the end-of-backup record will
8352 : * never arrive.
8353 : */
8354 70 : if (ArchiveRecoveryRequested &&
8355 68 : XLogRecPtrIsValid(ControlFile->backupStartPoint) &&
8356 0 : !XLogRecPtrIsValid(ControlFile->backupEndPoint))
8357 0 : ereport(PANIC,
8358 : (errmsg("online backup was canceled, recovery cannot continue")));
8359 :
8360 : /*
8361 : * If we see a shutdown checkpoint, we know that nothing was running
8362 : * on the primary at this point. So fake-up an empty running-xacts
8363 : * record and use that here and now. Recover additional standby state
8364 : * for prepared transactions.
8365 : */
8366 70 : if (standbyState >= STANDBY_INITIALIZED)
8367 : {
8368 : TransactionId *xids;
8369 : int nxids;
8370 : TransactionId oldestActiveXID;
8371 : TransactionId latestCompletedXid;
8372 : RunningTransactionsData running;
8373 :
8374 64 : oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8375 :
8376 : /* Update pg_subtrans entries for any prepared transactions */
8377 64 : StandbyRecoverPreparedTransactions();
8378 :
8379 : /*
8380 : * Construct a RunningTransactions snapshot representing a shut
8381 : * down server, with only prepared transactions still alive. We're
8382 : * never overflowed at this point because all subxids are listed
8383 : * with their parent prepared transactions.
8384 : */
8385 64 : running.xcnt = nxids;
8386 64 : running.subxcnt = 0;
8387 64 : running.subxid_status = SUBXIDS_IN_SUBTRANS;
8388 64 : running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
8389 64 : running.oldestRunningXid = oldestActiveXID;
8390 64 : latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
8391 64 : TransactionIdRetreat(latestCompletedXid);
8392 : Assert(TransactionIdIsNormal(latestCompletedXid));
8393 64 : running.latestCompletedXid = latestCompletedXid;
8394 64 : running.xids = xids;
8395 :
8396 64 : ProcArrayApplyRecoveryInfo(&running);
8397 : }
8398 :
8399 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8400 70 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8401 70 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8402 70 : LWLockRelease(ControlFileLock);
8403 :
8404 : /*
8405 : * We should've already switched to the new TLI before replaying this
8406 : * record.
8407 : */
8408 70 : (void) GetCurrentReplayRecPtr(&replayTLI);
8409 70 : if (checkPoint.ThisTimeLineID != replayTLI)
8410 0 : ereport(PANIC,
8411 : (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8412 : checkPoint.ThisTimeLineID, replayTLI)));
8413 :
8414 70 : RecoveryRestartPoint(&checkPoint, record);
8415 :
8416 : /*
8417 : * After replaying a checkpoint record, free all smgr objects.
8418 : * Otherwise we would never do so for dropped relations, as the
8419 : * startup does not process shared invalidation messages or call
8420 : * AtEOXact_SMgr().
8421 : */
8422 70 : smgrdestroyall();
8423 : }
8424 87320 : else if (info == XLOG_CHECKPOINT_ONLINE)
8425 : {
8426 : CheckPoint checkPoint;
8427 : TimeLineID replayTLI;
8428 :
8429 1356 : memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8430 : /* In an ONLINE checkpoint, treat the XID counter as a minimum */
8431 1356 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8432 1356 : if (FullTransactionIdPrecedes(TransamVariables->nextXid,
8433 : checkPoint.nextXid))
8434 0 : TransamVariables->nextXid = checkPoint.nextXid;
8435 1356 : LWLockRelease(XidGenLock);
8436 :
8437 : /*
8438 : * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8439 : * to track OID assignment through XLOG_NEXTOID records. The nextOid
8440 : * counter is from the start of the checkpoint and might well be stale
8441 : * compared to later XLOG_NEXTOID records. We could try to take the
8442 : * maximum of the nextOid counter and our latest value, but since
8443 : * there's no particular guarantee about the speed with which the OID
8444 : * counter wraps around, that's a risky thing to do. In any case,
8445 : * users of the nextOid counter are required to avoid assignment of
8446 : * duplicates, so that a somewhat out-of-date value should be safe.
8447 : */
8448 :
8449 : /* Handle multixact */
8450 1356 : MultiXactAdvanceNextMXact(checkPoint.nextMulti,
8451 : checkPoint.nextMultiOffset);
8452 :
8453 : /*
8454 : * NB: This may perform multixact truncation when replaying WAL
8455 : * generated by an older primary.
8456 : */
8457 1356 : MultiXactAdvanceOldest(checkPoint.oldestMulti,
8458 : checkPoint.oldestMultiDB);
8459 1356 : if (TransactionIdPrecedes(TransamVariables->oldestXid,
8460 : checkPoint.oldestXid))
8461 0 : SetTransactionIdLimit(checkPoint.oldestXid,
8462 : checkPoint.oldestXidDB);
8463 : /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8464 1356 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8465 1356 : ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8466 1356 : LWLockRelease(ControlFileLock);
8467 :
8468 : /* TLI should not change in an on-line checkpoint */
8469 1356 : (void) GetCurrentReplayRecPtr(&replayTLI);
8470 1356 : if (checkPoint.ThisTimeLineID != replayTLI)
8471 0 : ereport(PANIC,
8472 : (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8473 : checkPoint.ThisTimeLineID, replayTLI)));
8474 :
8475 1356 : RecoveryRestartPoint(&checkPoint, record);
8476 :
8477 : /*
8478 : * After replaying a checkpoint record, free all smgr objects.
8479 : * Otherwise we would never do so for dropped relations, as the
8480 : * startup does not process shared invalidation messages or call
8481 : * AtEOXact_SMgr().
8482 : */
8483 1356 : smgrdestroyall();
8484 : }
8485 85964 : else if (info == XLOG_OVERWRITE_CONTRECORD)
8486 : {
8487 : /* nothing to do here, handled in xlogrecovery_redo() */
8488 : }
8489 85962 : else if (info == XLOG_END_OF_RECOVERY)
8490 : {
8491 : xl_end_of_recovery xlrec;
8492 : TimeLineID replayTLI;
8493 :
8494 20 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8495 :
8496 : /*
8497 : * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8498 : * but this case is rarer and harder to test, so the benefit doesn't
8499 : * outweigh the potential extra cost of maintenance.
8500 : */
8501 :
8502 : /*
8503 : * We should've already switched to the new TLI before replaying this
8504 : * record.
8505 : */
8506 20 : (void) GetCurrentReplayRecPtr(&replayTLI);
8507 20 : if (xlrec.ThisTimeLineID != replayTLI)
8508 0 : ereport(PANIC,
8509 : (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8510 : xlrec.ThisTimeLineID, replayTLI)));
8511 : }
8512 85942 : else if (info == XLOG_NOOP)
8513 : {
8514 : /* nothing to do here */
8515 : }
8516 85942 : else if (info == XLOG_SWITCH)
8517 : {
8518 : /* nothing to do here */
8519 : }
8520 85048 : else if (info == XLOG_RESTORE_POINT)
8521 : {
8522 : /* nothing to do here, handled in xlogrecovery.c */
8523 : }
8524 85038 : else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8525 : {
8526 : /*
8527 : * XLOG_FPI records contain nothing else but one or more block
8528 : * references. Every block reference must include a full-page image
8529 : * even if full_page_writes was disabled when the record was generated
8530 : * - otherwise there would be no point in this record.
8531 : *
8532 : * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8533 : * WAL-logged because of a hint bit update. They are only generated
8534 : * when checksums and/or wal_log_hints are enabled. They may include
8535 : * no full-page images if full_page_writes was disabled when they were
8536 : * generated. In this case there is nothing to do here.
8537 : *
8538 : * No recovery conflicts are generated by these generic records - if a
8539 : * resource manager needs to generate conflicts, it has to define a
8540 : * separate WAL record type and redo routine.
8541 : */
8542 176316 : for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8543 : {
8544 : Buffer buffer;
8545 :
8546 92876 : if (!XLogRecHasBlockImage(record, block_id))
8547 : {
8548 132 : if (info == XLOG_FPI)
8549 0 : elog(ERROR, "XLOG_FPI record did not contain a full-page image");
8550 132 : continue;
8551 : }
8552 :
8553 92744 : if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
8554 0 : elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
8555 92744 : UnlockReleaseBuffer(buffer);
8556 : }
8557 : }
8558 1598 : else if (info == XLOG_BACKUP_END)
8559 : {
8560 : /* nothing to do here, handled in xlogrecovery_redo() */
8561 : }
8562 1424 : else if (info == XLOG_PARAMETER_CHANGE)
8563 : {
8564 : xl_parameter_change xlrec;
8565 :
8566 : /* Update our copy of the parameters in pg_control */
8567 66 : memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8568 :
8569 : /*
8570 : * Invalidate logical slots if we are in hot standby and the primary
8571 : * does not have a WAL level sufficient for logical decoding. No need
8572 : * to search for potentially conflicting logically slots if standby is
8573 : * running with wal_level lower than logical, because in that case, we
8574 : * would have either disallowed creation of logical slots or
8575 : * invalidated existing ones.
8576 : */
8577 66 : if (InRecovery && InHotStandby &&
8578 36 : xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8579 14 : wal_level >= WAL_LEVEL_LOGICAL)
8580 8 : InvalidateObsoleteReplicationSlots(RS_INVAL_WAL_LEVEL,
8581 : 0, InvalidOid,
8582 : InvalidTransactionId);
8583 :
8584 66 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8585 66 : ControlFile->MaxConnections = xlrec.MaxConnections;
8586 66 : ControlFile->max_worker_processes = xlrec.max_worker_processes;
8587 66 : ControlFile->max_wal_senders = xlrec.max_wal_senders;
8588 66 : ControlFile->max_prepared_xacts = xlrec.max_prepared_xacts;
8589 66 : ControlFile->max_locks_per_xact = xlrec.max_locks_per_xact;
8590 66 : ControlFile->wal_level = xlrec.wal_level;
8591 66 : ControlFile->wal_log_hints = xlrec.wal_log_hints;
8592 :
8593 : /*
8594 : * Update minRecoveryPoint to ensure that if recovery is aborted, we
8595 : * recover back up to this point before allowing hot standby again.
8596 : * This is important if the max_* settings are decreased, to ensure
8597 : * you don't run queries against the WAL preceding the change. The
8598 : * local copies cannot be updated as long as crash recovery is
8599 : * happening and we expect all the WAL to be replayed.
8600 : */
8601 66 : if (InArchiveRecovery)
8602 : {
8603 38 : LocalMinRecoveryPoint = ControlFile->minRecoveryPoint;
8604 38 : LocalMinRecoveryPointTLI = ControlFile->minRecoveryPointTLI;
8605 : }
8606 66 : if (XLogRecPtrIsValid(LocalMinRecoveryPoint) && LocalMinRecoveryPoint < lsn)
8607 : {
8608 : TimeLineID replayTLI;
8609 :
8610 16 : (void) GetCurrentReplayRecPtr(&replayTLI);
8611 16 : ControlFile->minRecoveryPoint = lsn;
8612 16 : ControlFile->minRecoveryPointTLI = replayTLI;
8613 : }
8614 :
8615 66 : CommitTsParameterChange(xlrec.track_commit_timestamp,
8616 66 : ControlFile->track_commit_timestamp);
8617 66 : ControlFile->track_commit_timestamp = xlrec.track_commit_timestamp;
8618 :
8619 66 : UpdateControlFile();
8620 66 : LWLockRelease(ControlFileLock);
8621 :
8622 : /* Check to see if any parameter change gives a problem on recovery */
8623 66 : CheckRequiredParameterValues();
8624 : }
8625 1358 : else if (info == XLOG_FPW_CHANGE)
8626 : {
8627 : bool fpw;
8628 :
8629 0 : memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8630 :
8631 : /*
8632 : * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8633 : * do_pg_backup_start() and do_pg_backup_stop() can check whether
8634 : * full_page_writes has been disabled during online backup.
8635 : */
8636 0 : if (!fpw)
8637 : {
8638 0 : SpinLockAcquire(&XLogCtl->info_lck);
8639 0 : if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8640 0 : XLogCtl->lastFpwDisableRecPtr = record->ReadRecPtr;
8641 0 : SpinLockRelease(&XLogCtl->info_lck);
8642 : }
8643 :
8644 : /* Keep track of full_page_writes */
8645 0 : lastFullPageWrites = fpw;
8646 : }
8647 : else if (info == XLOG_CHECKPOINT_REDO)
8648 : {
8649 : /* nothing to do here, just for informational purposes */
8650 : }
8651 87572 : }
8652 :
8653 : /*
8654 : * Return the extra open flags used for opening a file, depending on the
8655 : * value of the GUCs wal_sync_method, fsync and debug_io_direct.
8656 : */
8657 : static int
8658 31678 : get_sync_bit(int method)
8659 : {
8660 31678 : int o_direct_flag = 0;
8661 :
8662 : /*
8663 : * Use O_DIRECT if requested, except in walreceiver process. The WAL
8664 : * written by walreceiver is normally read by the startup process soon
8665 : * after it's written. Also, walreceiver performs unaligned writes, which
8666 : * don't work with O_DIRECT, so it is required for correctness too.
8667 : */
8668 31678 : if ((io_direct_flags & IO_DIRECT_WAL) && !AmWalReceiverProcess())
8669 16 : o_direct_flag = PG_O_DIRECT;
8670 :
8671 : /* If fsync is disabled, never open in sync mode */
8672 31678 : if (!enableFsync)
8673 31678 : return o_direct_flag;
8674 :
8675 0 : switch (method)
8676 : {
8677 : /*
8678 : * enum values for all sync options are defined even if they are
8679 : * not supported on the current platform. But if not, they are
8680 : * not included in the enum option array, and therefore will never
8681 : * be seen here.
8682 : */
8683 0 : case WAL_SYNC_METHOD_FSYNC:
8684 : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8685 : case WAL_SYNC_METHOD_FDATASYNC:
8686 0 : return o_direct_flag;
8687 : #ifdef O_SYNC
8688 0 : case WAL_SYNC_METHOD_OPEN:
8689 0 : return O_SYNC | o_direct_flag;
8690 : #endif
8691 : #ifdef O_DSYNC
8692 0 : case WAL_SYNC_METHOD_OPEN_DSYNC:
8693 0 : return O_DSYNC | o_direct_flag;
8694 : #endif
8695 0 : default:
8696 : /* can't happen (unless we are out of sync with option array) */
8697 0 : elog(ERROR, "unrecognized \"wal_sync_method\": %d", method);
8698 : return 0; /* silence warning */
8699 : }
8700 : }
8701 :
8702 : /*
8703 : * GUC support
8704 : */
8705 : void
8706 2272 : assign_wal_sync_method(int new_wal_sync_method, void *extra)
8707 : {
8708 2272 : if (wal_sync_method != new_wal_sync_method)
8709 : {
8710 : /*
8711 : * To ensure that no blocks escape unsynced, force an fsync on the
8712 : * currently open log segment (if any). Also, if the open flag is
8713 : * changing, close the log file so it will be reopened (with new flag
8714 : * bit) at next use.
8715 : */
8716 0 : if (openLogFile >= 0)
8717 : {
8718 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
8719 0 : if (pg_fsync(openLogFile) != 0)
8720 : {
8721 : char xlogfname[MAXFNAMELEN];
8722 : int save_errno;
8723 :
8724 0 : save_errno = errno;
8725 0 : XLogFileName(xlogfname, openLogTLI, openLogSegNo,
8726 : wal_segment_size);
8727 0 : errno = save_errno;
8728 0 : ereport(PANIC,
8729 : (errcode_for_file_access(),
8730 : errmsg("could not fsync file \"%s\": %m", xlogfname)));
8731 : }
8732 :
8733 0 : pgstat_report_wait_end();
8734 0 : if (get_sync_bit(wal_sync_method) != get_sync_bit(new_wal_sync_method))
8735 0 : XLogFileClose();
8736 : }
8737 : }
8738 2272 : }
8739 :
8740 :
8741 : /*
8742 : * Issue appropriate kind of fsync (if any) for an XLOG output file.
8743 : *
8744 : * 'fd' is a file descriptor for the XLOG file to be fsync'd.
8745 : * 'segno' is for error reporting purposes.
8746 : */
8747 : void
8748 307222 : issue_xlog_fsync(int fd, XLogSegNo segno, TimeLineID tli)
8749 : {
8750 307222 : char *msg = NULL;
8751 : instr_time start;
8752 :
8753 : Assert(tli != 0);
8754 :
8755 : /*
8756 : * Quick exit if fsync is disabled or write() has already synced the WAL
8757 : * file.
8758 : */
8759 307222 : if (!enableFsync ||
8760 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN ||
8761 0 : wal_sync_method == WAL_SYNC_METHOD_OPEN_DSYNC)
8762 307222 : return;
8763 :
8764 : /*
8765 : * Measure I/O timing to sync the WAL file for pg_stat_io.
8766 : */
8767 0 : start = pgstat_prepare_io_time(track_wal_io_timing);
8768 :
8769 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
8770 0 : switch (wal_sync_method)
8771 : {
8772 0 : case WAL_SYNC_METHOD_FSYNC:
8773 0 : if (pg_fsync_no_writethrough(fd) != 0)
8774 0 : msg = _("could not fsync file \"%s\": %m");
8775 0 : break;
8776 : #ifdef HAVE_FSYNC_WRITETHROUGH
8777 : case WAL_SYNC_METHOD_FSYNC_WRITETHROUGH:
8778 : if (pg_fsync_writethrough(fd) != 0)
8779 : msg = _("could not fsync write-through file \"%s\": %m");
8780 : break;
8781 : #endif
8782 0 : case WAL_SYNC_METHOD_FDATASYNC:
8783 0 : if (pg_fdatasync(fd) != 0)
8784 0 : msg = _("could not fdatasync file \"%s\": %m");
8785 0 : break;
8786 0 : case WAL_SYNC_METHOD_OPEN:
8787 : case WAL_SYNC_METHOD_OPEN_DSYNC:
8788 : /* not reachable */
8789 : Assert(false);
8790 0 : break;
8791 0 : default:
8792 0 : ereport(PANIC,
8793 : errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8794 : errmsg_internal("unrecognized \"wal_sync_method\": %d", wal_sync_method));
8795 : break;
8796 : }
8797 :
8798 : /* PANIC if failed to fsync */
8799 0 : if (msg)
8800 : {
8801 : char xlogfname[MAXFNAMELEN];
8802 0 : int save_errno = errno;
8803 :
8804 0 : XLogFileName(xlogfname, tli, segno, wal_segment_size);
8805 0 : errno = save_errno;
8806 0 : ereport(PANIC,
8807 : (errcode_for_file_access(),
8808 : errmsg(msg, xlogfname)));
8809 : }
8810 :
8811 0 : pgstat_report_wait_end();
8812 :
8813 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL, IOOP_FSYNC,
8814 : start, 1, 0);
8815 : }
8816 :
8817 : /*
8818 : * do_pg_backup_start is the workhorse of the user-visible pg_backup_start()
8819 : * function. It creates the necessary starting checkpoint and constructs the
8820 : * backup state and tablespace map.
8821 : *
8822 : * Input parameters are "state" (the backup state), "fast" (if true, we do
8823 : * the checkpoint in fast mode), and "tablespaces" (if non-NULL, indicates a
8824 : * list of tablespaceinfo structs describing the cluster's tablespaces.).
8825 : *
8826 : * The tablespace map contents are appended to passed-in parameter
8827 : * tablespace_map and the caller is responsible for including it in the backup
8828 : * archive as 'tablespace_map'. The tablespace_map file is required mainly for
8829 : * tar format in windows as native windows utilities are not able to create
8830 : * symlinks while extracting files from tar. However for consistency and
8831 : * platform-independence, we do it the same way everywhere.
8832 : *
8833 : * It fills in "state" with the information required for the backup, such
8834 : * as the minimum WAL location that must be present to restore from this
8835 : * backup (starttli) and the corresponding timeline ID (starttli).
8836 : *
8837 : * Every successfully started backup must be stopped by calling
8838 : * do_pg_backup_stop() or do_pg_abort_backup(). There can be many
8839 : * backups active at the same time.
8840 : *
8841 : * It is the responsibility of the caller of this function to verify the
8842 : * permissions of the calling user!
8843 : */
8844 : void
8845 328 : do_pg_backup_start(const char *backupidstr, bool fast, List **tablespaces,
8846 : BackupState *state, StringInfo tblspcmapfile)
8847 : {
8848 : bool backup_started_in_recovery;
8849 :
8850 : Assert(state != NULL);
8851 328 : backup_started_in_recovery = RecoveryInProgress();
8852 :
8853 : /*
8854 : * During recovery, we don't need to check WAL level. Because, if WAL
8855 : * level is not sufficient, it's impossible to get here during recovery.
8856 : */
8857 328 : if (!backup_started_in_recovery && !XLogIsNeeded())
8858 0 : ereport(ERROR,
8859 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8860 : errmsg("WAL level not sufficient for making an online backup"),
8861 : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
8862 :
8863 328 : if (strlen(backupidstr) > MAXPGPATH)
8864 2 : ereport(ERROR,
8865 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8866 : errmsg("backup label too long (max %d bytes)",
8867 : MAXPGPATH)));
8868 :
8869 326 : strlcpy(state->name, backupidstr, sizeof(state->name));
8870 :
8871 : /*
8872 : * Mark backup active in shared memory. We must do full-page WAL writes
8873 : * during an on-line backup even if not doing so at other times, because
8874 : * it's quite possible for the backup dump to obtain a "torn" (partially
8875 : * written) copy of a database page if it reads the page concurrently with
8876 : * our write to the same page. This can be fixed as long as the first
8877 : * write to the page in the WAL sequence is a full-page write. Hence, we
8878 : * increment runningBackups then force a CHECKPOINT, to ensure there are
8879 : * no dirty pages in shared memory that might get dumped while the backup
8880 : * is in progress without having a corresponding WAL record. (Once the
8881 : * backup is complete, we need not force full-page writes anymore, since
8882 : * we expect that any pages not modified during the backup interval must
8883 : * have been correctly captured by the backup.)
8884 : *
8885 : * Note that forcing full-page writes has no effect during an online
8886 : * backup from the standby.
8887 : *
8888 : * We must hold all the insertion locks to change the value of
8889 : * runningBackups, to ensure adequate interlocking against
8890 : * XLogInsertRecord().
8891 : */
8892 326 : WALInsertLockAcquireExclusive();
8893 326 : XLogCtl->Insert.runningBackups++;
8894 326 : WALInsertLockRelease();
8895 :
8896 : /*
8897 : * Ensure we decrement runningBackups if we fail below. NB -- for this to
8898 : * work correctly, it is critical that sessionBackupState is only updated
8899 : * after this block is over.
8900 : */
8901 326 : PG_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
8902 : {
8903 326 : bool gotUniqueStartpoint = false;
8904 : DIR *tblspcdir;
8905 : struct dirent *de;
8906 : tablespaceinfo *ti;
8907 : int datadirpathlen;
8908 :
8909 : /*
8910 : * Force an XLOG file switch before the checkpoint, to ensure that the
8911 : * WAL segment the checkpoint is written to doesn't contain pages with
8912 : * old timeline IDs. That would otherwise happen if you called
8913 : * pg_backup_start() right after restoring from a PITR archive: the
8914 : * first WAL segment containing the startup checkpoint has pages in
8915 : * the beginning with the old timeline ID. That can cause trouble at
8916 : * recovery: we won't have a history file covering the old timeline if
8917 : * pg_wal directory was not included in the base backup and the WAL
8918 : * archive was cleared too before starting the backup.
8919 : *
8920 : * This also ensures that we have emitted a WAL page header that has
8921 : * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8922 : * Therefore, if a WAL archiver (such as pglesslog) is trying to
8923 : * compress out removable backup blocks, it won't remove any that
8924 : * occur after this point.
8925 : *
8926 : * During recovery, we skip forcing XLOG file switch, which means that
8927 : * the backup taken during recovery is not available for the special
8928 : * recovery case described above.
8929 : */
8930 326 : if (!backup_started_in_recovery)
8931 312 : RequestXLogSwitch(false);
8932 :
8933 : do
8934 : {
8935 : bool checkpointfpw;
8936 :
8937 : /*
8938 : * Force a CHECKPOINT. Aside from being necessary to prevent torn
8939 : * page problems, this guarantees that two successive backup runs
8940 : * will have different checkpoint positions and hence different
8941 : * history file names, even if nothing happened in between.
8942 : *
8943 : * During recovery, establish a restartpoint if possible. We use
8944 : * the last restartpoint as the backup starting checkpoint. This
8945 : * means that two successive backup runs can have same checkpoint
8946 : * positions.
8947 : *
8948 : * Since the fact that we are executing do_pg_backup_start()
8949 : * during recovery means that checkpointer is running, we can use
8950 : * RequestCheckpoint() to establish a restartpoint.
8951 : *
8952 : * We use CHECKPOINT_FAST only if requested by user (via passing
8953 : * fast = true). Otherwise this can take awhile.
8954 : */
8955 326 : RequestCheckpoint(CHECKPOINT_FORCE | CHECKPOINT_WAIT |
8956 : (fast ? CHECKPOINT_FAST : 0));
8957 :
8958 : /*
8959 : * Now we need to fetch the checkpoint record location, and also
8960 : * its REDO pointer. The oldest point in WAL that would be needed
8961 : * to restore starting from the checkpoint is precisely the REDO
8962 : * pointer.
8963 : */
8964 326 : LWLockAcquire(ControlFileLock, LW_SHARED);
8965 326 : state->checkpointloc = ControlFile->checkPoint;
8966 326 : state->startpoint = ControlFile->checkPointCopy.redo;
8967 326 : state->starttli = ControlFile->checkPointCopy.ThisTimeLineID;
8968 326 : checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
8969 326 : LWLockRelease(ControlFileLock);
8970 :
8971 326 : if (backup_started_in_recovery)
8972 : {
8973 : XLogRecPtr recptr;
8974 :
8975 : /*
8976 : * Check to see if all WAL replayed during online backup
8977 : * (i.e., since last restartpoint used as backup starting
8978 : * checkpoint) contain full-page writes.
8979 : */
8980 14 : SpinLockAcquire(&XLogCtl->info_lck);
8981 14 : recptr = XLogCtl->lastFpwDisableRecPtr;
8982 14 : SpinLockRelease(&XLogCtl->info_lck);
8983 :
8984 14 : if (!checkpointfpw || state->startpoint <= recptr)
8985 0 : ereport(ERROR,
8986 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8987 : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
8988 : "since last restartpoint"),
8989 : errhint("This means that the backup being taken on the standby "
8990 : "is corrupt and should not be used. "
8991 : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
8992 : "and then try an online backup again.")));
8993 :
8994 : /*
8995 : * During recovery, since we don't use the end-of-backup WAL
8996 : * record and don't write the backup history file, the
8997 : * starting WAL location doesn't need to be unique. This means
8998 : * that two base backups started at the same time might use
8999 : * the same checkpoint as starting locations.
9000 : */
9001 14 : gotUniqueStartpoint = true;
9002 : }
9003 :
9004 : /*
9005 : * If two base backups are started at the same time (in WAL sender
9006 : * processes), we need to make sure that they use different
9007 : * checkpoints as starting locations, because we use the starting
9008 : * WAL location as a unique identifier for the base backup in the
9009 : * end-of-backup WAL record and when we write the backup history
9010 : * file. Perhaps it would be better generate a separate unique ID
9011 : * for each backup instead of forcing another checkpoint, but
9012 : * taking a checkpoint right after another is not that expensive
9013 : * either because only few buffers have been dirtied yet.
9014 : */
9015 326 : WALInsertLockAcquireExclusive();
9016 326 : if (XLogCtl->Insert.lastBackupStart < state->startpoint)
9017 : {
9018 326 : XLogCtl->Insert.lastBackupStart = state->startpoint;
9019 326 : gotUniqueStartpoint = true;
9020 : }
9021 326 : WALInsertLockRelease();
9022 326 : } while (!gotUniqueStartpoint);
9023 :
9024 : /*
9025 : * Construct tablespace_map file.
9026 : */
9027 326 : datadirpathlen = strlen(DataDir);
9028 :
9029 : /* Collect information about all tablespaces */
9030 326 : tblspcdir = AllocateDir(PG_TBLSPC_DIR);
9031 1050 : while ((de = ReadDir(tblspcdir, PG_TBLSPC_DIR)) != NULL)
9032 : {
9033 : char fullpath[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
9034 : char linkpath[MAXPGPATH];
9035 724 : char *relpath = NULL;
9036 : char *s;
9037 : PGFileType de_type;
9038 : char *badp;
9039 : Oid tsoid;
9040 :
9041 : /*
9042 : * Try to parse the directory name as an unsigned integer.
9043 : *
9044 : * Tablespace directories should be positive integers that can be
9045 : * represented in 32 bits, with no leading zeroes or trailing
9046 : * garbage. If we come across a name that doesn't meet those
9047 : * criteria, skip it.
9048 : */
9049 724 : if (de->d_name[0] < '1' || de->d_name[1] > '9')
9050 652 : continue;
9051 72 : errno = 0;
9052 72 : tsoid = strtoul(de->d_name, &badp, 10);
9053 72 : if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
9054 0 : continue;
9055 :
9056 72 : snprintf(fullpath, sizeof(fullpath), "%s/%s", PG_TBLSPC_DIR, de->d_name);
9057 :
9058 72 : de_type = get_dirent_type(fullpath, de, false, ERROR);
9059 :
9060 72 : if (de_type == PGFILETYPE_LNK)
9061 : {
9062 : StringInfoData escapedpath;
9063 : int rllen;
9064 :
9065 44 : rllen = readlink(fullpath, linkpath, sizeof(linkpath));
9066 44 : if (rllen < 0)
9067 : {
9068 0 : ereport(WARNING,
9069 : (errmsg("could not read symbolic link \"%s\": %m",
9070 : fullpath)));
9071 0 : continue;
9072 : }
9073 44 : else if (rllen >= sizeof(linkpath))
9074 : {
9075 0 : ereport(WARNING,
9076 : (errmsg("symbolic link \"%s\" target is too long",
9077 : fullpath)));
9078 0 : continue;
9079 : }
9080 44 : linkpath[rllen] = '\0';
9081 :
9082 : /*
9083 : * Relpath holds the relative path of the tablespace directory
9084 : * when it's located within PGDATA, or NULL if it's located
9085 : * elsewhere.
9086 : */
9087 44 : if (rllen > datadirpathlen &&
9088 2 : strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
9089 0 : IS_DIR_SEP(linkpath[datadirpathlen]))
9090 0 : relpath = pstrdup(linkpath + datadirpathlen + 1);
9091 :
9092 : /*
9093 : * Add a backslash-escaped version of the link path to the
9094 : * tablespace map file.
9095 : */
9096 44 : initStringInfo(&escapedpath);
9097 1092 : for (s = linkpath; *s; s++)
9098 : {
9099 1048 : if (*s == '\n' || *s == '\r' || *s == '\\')
9100 0 : appendStringInfoChar(&escapedpath, '\\');
9101 1048 : appendStringInfoChar(&escapedpath, *s);
9102 : }
9103 44 : appendStringInfo(tblspcmapfile, "%s %s\n",
9104 44 : de->d_name, escapedpath.data);
9105 44 : pfree(escapedpath.data);
9106 : }
9107 28 : else if (de_type == PGFILETYPE_DIR)
9108 : {
9109 : /*
9110 : * It's possible to use allow_in_place_tablespaces to create
9111 : * directories directly under pg_tblspc, for testing purposes
9112 : * only.
9113 : *
9114 : * In this case, we store a relative path rather than an
9115 : * absolute path into the tablespaceinfo.
9116 : */
9117 28 : snprintf(linkpath, sizeof(linkpath), "%s/%s",
9118 28 : PG_TBLSPC_DIR, de->d_name);
9119 28 : relpath = pstrdup(linkpath);
9120 : }
9121 : else
9122 : {
9123 : /* Skip any other file type that appears here. */
9124 0 : continue;
9125 : }
9126 :
9127 72 : ti = palloc(sizeof(tablespaceinfo));
9128 72 : ti->oid = tsoid;
9129 72 : ti->path = pstrdup(linkpath);
9130 72 : ti->rpath = relpath;
9131 72 : ti->size = -1;
9132 :
9133 72 : if (tablespaces)
9134 72 : *tablespaces = lappend(*tablespaces, ti);
9135 : }
9136 326 : FreeDir(tblspcdir);
9137 :
9138 326 : state->starttime = (pg_time_t) time(NULL);
9139 : }
9140 326 : PG_END_ENSURE_ERROR_CLEANUP(do_pg_abort_backup, BoolGetDatum(true));
9141 :
9142 326 : state->started_in_recovery = backup_started_in_recovery;
9143 :
9144 : /*
9145 : * Mark that the start phase has correctly finished for the backup.
9146 : */
9147 326 : sessionBackupState = SESSION_BACKUP_RUNNING;
9148 326 : }
9149 :
9150 : /*
9151 : * Utility routine to fetch the session-level status of a backup running.
9152 : */
9153 : SessionBackupState
9154 368 : get_backup_status(void)
9155 : {
9156 368 : return sessionBackupState;
9157 : }
9158 :
9159 : /*
9160 : * do_pg_backup_stop
9161 : *
9162 : * Utility function called at the end of an online backup. It creates history
9163 : * file (if required), resets sessionBackupState and so on. It can optionally
9164 : * wait for WAL segments to be archived.
9165 : *
9166 : * "state" is filled with the information necessary to restore from this
9167 : * backup with its stop LSN (stoppoint), its timeline ID (stoptli), etc.
9168 : *
9169 : * It is the responsibility of the caller of this function to verify the
9170 : * permissions of the calling user!
9171 : */
9172 : void
9173 314 : do_pg_backup_stop(BackupState *state, bool waitforarchive)
9174 : {
9175 314 : bool backup_stopped_in_recovery = false;
9176 : char histfilepath[MAXPGPATH];
9177 : char lastxlogfilename[MAXFNAMELEN];
9178 : char histfilename[MAXFNAMELEN];
9179 : XLogSegNo _logSegNo;
9180 : FILE *fp;
9181 : int seconds_before_warning;
9182 314 : int waits = 0;
9183 314 : bool reported_waiting = false;
9184 :
9185 : Assert(state != NULL);
9186 :
9187 314 : backup_stopped_in_recovery = RecoveryInProgress();
9188 :
9189 : /*
9190 : * During recovery, we don't need to check WAL level. Because, if WAL
9191 : * level is not sufficient, it's impossible to get here during recovery.
9192 : */
9193 314 : if (!backup_stopped_in_recovery && !XLogIsNeeded())
9194 0 : ereport(ERROR,
9195 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9196 : errmsg("WAL level not sufficient for making an online backup"),
9197 : errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
9198 :
9199 : /*
9200 : * OK to update backup counter and session-level lock.
9201 : *
9202 : * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
9203 : * otherwise they can be updated inconsistently, which might cause
9204 : * do_pg_abort_backup() to fail.
9205 : */
9206 314 : WALInsertLockAcquireExclusive();
9207 :
9208 : /*
9209 : * It is expected that each do_pg_backup_start() call is matched by
9210 : * exactly one do_pg_backup_stop() call.
9211 : */
9212 : Assert(XLogCtl->Insert.runningBackups > 0);
9213 314 : XLogCtl->Insert.runningBackups--;
9214 :
9215 : /*
9216 : * Clean up session-level lock.
9217 : *
9218 : * You might think that WALInsertLockRelease() can be called before
9219 : * cleaning up session-level lock because session-level lock doesn't need
9220 : * to be protected with WAL insertion lock. But since
9221 : * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9222 : * cleaned up before it.
9223 : */
9224 314 : sessionBackupState = SESSION_BACKUP_NONE;
9225 :
9226 314 : WALInsertLockRelease();
9227 :
9228 : /*
9229 : * If we are taking an online backup from the standby, we confirm that the
9230 : * standby has not been promoted during the backup.
9231 : */
9232 314 : if (state->started_in_recovery && !backup_stopped_in_recovery)
9233 0 : ereport(ERROR,
9234 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9235 : errmsg("the standby was promoted during online backup"),
9236 : errhint("This means that the backup being taken is corrupt "
9237 : "and should not be used. "
9238 : "Try taking another online backup.")));
9239 :
9240 : /*
9241 : * During recovery, we don't write an end-of-backup record. We assume that
9242 : * pg_control was backed up last and its minimum recovery point can be
9243 : * available as the backup end location. Since we don't have an
9244 : * end-of-backup record, we use the pg_control value to check whether
9245 : * we've reached the end of backup when starting recovery from this
9246 : * backup. We have no way of checking if pg_control wasn't backed up last
9247 : * however.
9248 : *
9249 : * We don't force a switch to new WAL file but it is still possible to
9250 : * wait for all the required files to be archived if waitforarchive is
9251 : * true. This is okay if we use the backup to start a standby and fetch
9252 : * the missing WAL using streaming replication. But in the case of an
9253 : * archive recovery, a user should set waitforarchive to true and wait for
9254 : * them to be archived to ensure that all the required files are
9255 : * available.
9256 : *
9257 : * We return the current minimum recovery point as the backup end
9258 : * location. Note that it can be greater than the exact backup end
9259 : * location if the minimum recovery point is updated after the backup of
9260 : * pg_control. This is harmless for current uses.
9261 : *
9262 : * XXX currently a backup history file is for informational and debug
9263 : * purposes only. It's not essential for an online backup. Furthermore,
9264 : * even if it's created, it will not be archived during recovery because
9265 : * an archiver is not invoked. So it doesn't seem worthwhile to write a
9266 : * backup history file during recovery.
9267 : */
9268 314 : if (backup_stopped_in_recovery)
9269 : {
9270 : XLogRecPtr recptr;
9271 :
9272 : /*
9273 : * Check to see if all WAL replayed during online backup contain
9274 : * full-page writes.
9275 : */
9276 14 : SpinLockAcquire(&XLogCtl->info_lck);
9277 14 : recptr = XLogCtl->lastFpwDisableRecPtr;
9278 14 : SpinLockRelease(&XLogCtl->info_lck);
9279 :
9280 14 : if (state->startpoint <= recptr)
9281 0 : ereport(ERROR,
9282 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9283 : errmsg("WAL generated with \"full_page_writes=off\" was replayed "
9284 : "during online backup"),
9285 : errhint("This means that the backup being taken on the standby "
9286 : "is corrupt and should not be used. "
9287 : "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
9288 : "and then try an online backup again.")));
9289 :
9290 :
9291 14 : LWLockAcquire(ControlFileLock, LW_SHARED);
9292 14 : state->stoppoint = ControlFile->minRecoveryPoint;
9293 14 : state->stoptli = ControlFile->minRecoveryPointTLI;
9294 14 : LWLockRelease(ControlFileLock);
9295 : }
9296 : else
9297 : {
9298 : char *history_file;
9299 :
9300 : /*
9301 : * Write the backup-end xlog record
9302 : */
9303 300 : XLogBeginInsert();
9304 300 : XLogRegisterData(&state->startpoint,
9305 : sizeof(state->startpoint));
9306 300 : state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9307 :
9308 : /*
9309 : * Given that we're not in recovery, InsertTimeLineID is set and can't
9310 : * change, so we can read it without a lock.
9311 : */
9312 300 : state->stoptli = XLogCtl->InsertTimeLineID;
9313 :
9314 : /*
9315 : * Force a switch to a new xlog segment file, so that the backup is
9316 : * valid as soon as archiver moves out the current segment file.
9317 : */
9318 300 : RequestXLogSwitch(false);
9319 :
9320 300 : state->stoptime = (pg_time_t) time(NULL);
9321 :
9322 : /*
9323 : * Write the backup history file
9324 : */
9325 300 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9326 300 : BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9327 : state->startpoint, wal_segment_size);
9328 300 : fp = AllocateFile(histfilepath, "w");
9329 300 : if (!fp)
9330 0 : ereport(ERROR,
9331 : (errcode_for_file_access(),
9332 : errmsg("could not create file \"%s\": %m",
9333 : histfilepath)));
9334 :
9335 : /* Build and save the contents of the backup history file */
9336 300 : history_file = build_backup_content(state, true);
9337 300 : fprintf(fp, "%s", history_file);
9338 300 : pfree(history_file);
9339 :
9340 300 : if (fflush(fp) || ferror(fp) || FreeFile(fp))
9341 0 : ereport(ERROR,
9342 : (errcode_for_file_access(),
9343 : errmsg("could not write file \"%s\": %m",
9344 : histfilepath)));
9345 :
9346 : /*
9347 : * Clean out any no-longer-needed history files. As a side effect,
9348 : * this will post a .ready file for the newly created history file,
9349 : * notifying the archiver that history file may be archived
9350 : * immediately.
9351 : */
9352 300 : CleanupBackupHistory();
9353 : }
9354 :
9355 : /*
9356 : * If archiving is enabled, wait for all the required WAL files to be
9357 : * archived before returning. If archiving isn't enabled, the required WAL
9358 : * needs to be transported via streaming replication (hopefully with
9359 : * wal_keep_size set high enough), or some more exotic mechanism like
9360 : * polling and copying files from pg_wal with script. We have no knowledge
9361 : * of those mechanisms, so it's up to the user to ensure that he gets all
9362 : * the required WAL.
9363 : *
9364 : * We wait until both the last WAL file filled during backup and the
9365 : * history file have been archived, and assume that the alphabetic sorting
9366 : * property of the WAL files ensures any earlier WAL files are safely
9367 : * archived as well.
9368 : *
9369 : * We wait forever, since archive_command is supposed to work and we
9370 : * assume the admin wanted his backup to work completely. If you don't
9371 : * wish to wait, then either waitforarchive should be passed in as false,
9372 : * or you can set statement_timeout. Also, some notices are issued to
9373 : * clue in anyone who might be doing this interactively.
9374 : */
9375 :
9376 314 : if (waitforarchive &&
9377 20 : ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
9378 2 : (backup_stopped_in_recovery && XLogArchivingAlways())))
9379 : {
9380 8 : XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9381 8 : XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9382 : wal_segment_size);
9383 :
9384 8 : XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9385 8 : BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9386 : state->startpoint, wal_segment_size);
9387 :
9388 8 : seconds_before_warning = 60;
9389 8 : waits = 0;
9390 :
9391 24 : while (XLogArchiveIsBusy(lastxlogfilename) ||
9392 8 : XLogArchiveIsBusy(histfilename))
9393 : {
9394 8 : CHECK_FOR_INTERRUPTS();
9395 :
9396 8 : if (!reported_waiting && waits > 5)
9397 : {
9398 0 : ereport(NOTICE,
9399 : (errmsg("base backup done, waiting for required WAL segments to be archived")));
9400 0 : reported_waiting = true;
9401 : }
9402 :
9403 8 : (void) WaitLatch(MyLatch,
9404 : WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
9405 : 1000L,
9406 : WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9407 8 : ResetLatch(MyLatch);
9408 :
9409 8 : if (++waits >= seconds_before_warning)
9410 : {
9411 0 : seconds_before_warning *= 2; /* This wraps in >10 years... */
9412 0 : ereport(WARNING,
9413 : (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9414 : waits),
9415 : errhint("Check that your \"archive_command\" is executing properly. "
9416 : "You can safely cancel this backup, "
9417 : "but the database backup will not be usable without all the WAL segments.")));
9418 : }
9419 : }
9420 :
9421 8 : ereport(NOTICE,
9422 : (errmsg("all required WAL segments have been archived")));
9423 : }
9424 306 : else if (waitforarchive)
9425 12 : ereport(NOTICE,
9426 : (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
9427 314 : }
9428 :
9429 :
9430 : /*
9431 : * do_pg_abort_backup: abort a running backup
9432 : *
9433 : * This does just the most basic steps of do_pg_backup_stop(), by taking the
9434 : * system out of backup mode, thus making it a lot more safe to call from
9435 : * an error handler.
9436 : *
9437 : * 'arg' indicates that it's being called during backup setup; so
9438 : * sessionBackupState has not been modified yet, but runningBackups has
9439 : * already been incremented. When it's false, then it's invoked as a
9440 : * before_shmem_exit handler, and therefore we must not change state
9441 : * unless sessionBackupState indicates that a backup is actually running.
9442 : *
9443 : * NB: This gets used as a PG_ENSURE_ERROR_CLEANUP callback and
9444 : * before_shmem_exit handler, hence the odd-looking signature.
9445 : */
9446 : void
9447 16 : do_pg_abort_backup(int code, Datum arg)
9448 : {
9449 16 : bool during_backup_start = DatumGetBool(arg);
9450 :
9451 : /* If called during backup start, there shouldn't be one already running */
9452 : Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9453 :
9454 16 : if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9455 : {
9456 12 : WALInsertLockAcquireExclusive();
9457 : Assert(XLogCtl->Insert.runningBackups > 0);
9458 12 : XLogCtl->Insert.runningBackups--;
9459 :
9460 12 : sessionBackupState = SESSION_BACKUP_NONE;
9461 12 : WALInsertLockRelease();
9462 :
9463 12 : if (!during_backup_start)
9464 12 : ereport(WARNING,
9465 : errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9466 : }
9467 16 : }
9468 :
9469 : /*
9470 : * Register a handler that will warn about unterminated backups at end of
9471 : * session, unless this has already been done.
9472 : */
9473 : void
9474 8 : register_persistent_abort_backup_handler(void)
9475 : {
9476 : static bool already_done = false;
9477 :
9478 8 : if (already_done)
9479 2 : return;
9480 6 : before_shmem_exit(do_pg_abort_backup, BoolGetDatum(false));
9481 6 : already_done = true;
9482 : }
9483 :
9484 : /*
9485 : * Get latest WAL insert pointer
9486 : */
9487 : XLogRecPtr
9488 3988 : GetXLogInsertRecPtr(void)
9489 : {
9490 3988 : XLogCtlInsert *Insert = &XLogCtl->Insert;
9491 : uint64 current_bytepos;
9492 :
9493 3988 : SpinLockAcquire(&Insert->insertpos_lck);
9494 3988 : current_bytepos = Insert->CurrBytePos;
9495 3988 : SpinLockRelease(&Insert->insertpos_lck);
9496 :
9497 3988 : return XLogBytePosToRecPtr(current_bytepos);
9498 : }
9499 :
9500 : /*
9501 : * Get latest WAL write pointer
9502 : */
9503 : XLogRecPtr
9504 18070 : GetXLogWriteRecPtr(void)
9505 : {
9506 18070 : RefreshXLogWriteResult(LogwrtResult);
9507 :
9508 18070 : return LogwrtResult.Write;
9509 : }
9510 :
9511 : /*
9512 : * Returns the redo pointer of the last checkpoint or restartpoint. This is
9513 : * the oldest point in WAL that we still need, if we have to restart recovery.
9514 : */
9515 : void
9516 784 : GetOldestRestartPoint(XLogRecPtr *oldrecptr, TimeLineID *oldtli)
9517 : {
9518 784 : LWLockAcquire(ControlFileLock, LW_SHARED);
9519 784 : *oldrecptr = ControlFile->checkPointCopy.redo;
9520 784 : *oldtli = ControlFile->checkPointCopy.ThisTimeLineID;
9521 784 : LWLockRelease(ControlFileLock);
9522 784 : }
9523 :
9524 : /* Thin wrapper around ShutdownWalRcv(). */
9525 : void
9526 1894 : XLogShutdownWalRcv(void)
9527 : {
9528 : Assert(AmStartupProcess() || !IsUnderPostmaster);
9529 :
9530 1894 : ShutdownWalRcv();
9531 1894 : ResetInstallXLogFileSegmentActive();
9532 1894 : }
9533 :
9534 : /* Enable WAL file recycling and preallocation. */
9535 : void
9536 2246 : SetInstallXLogFileSegmentActive(void)
9537 : {
9538 2246 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9539 2246 : XLogCtl->InstallXLogFileSegmentActive = true;
9540 2246 : LWLockRelease(ControlFileLock);
9541 2246 : }
9542 :
9543 : /* Disable WAL file recycling and preallocation. */
9544 : void
9545 2146 : ResetInstallXLogFileSegmentActive(void)
9546 : {
9547 2146 : LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9548 2146 : XLogCtl->InstallXLogFileSegmentActive = false;
9549 2146 : LWLockRelease(ControlFileLock);
9550 2146 : }
9551 :
9552 : bool
9553 0 : IsInstallXLogFileSegmentActive(void)
9554 : {
9555 : bool result;
9556 :
9557 0 : LWLockAcquire(ControlFileLock, LW_SHARED);
9558 0 : result = XLogCtl->InstallXLogFileSegmentActive;
9559 0 : LWLockRelease(ControlFileLock);
9560 :
9561 0 : return result;
9562 : }
9563 :
9564 : /*
9565 : * Update the WalWriterSleeping flag.
9566 : */
9567 : void
9568 1010 : SetWalWriterSleeping(bool sleeping)
9569 : {
9570 1010 : SpinLockAcquire(&XLogCtl->info_lck);
9571 1010 : XLogCtl->WalWriterSleeping = sleeping;
9572 1010 : SpinLockRelease(&XLogCtl->info_lck);
9573 1010 : }
|
