PostgreSQL Source Code  git master
slru.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * slru.c
4  * Simple LRU buffering for transaction status logfiles
5  *
6  * We use a simple least-recently-used scheme to manage a pool of page
7  * buffers. Under ordinary circumstances we expect that write
8  * traffic will occur mostly to the latest page (and to the just-prior
9  * page, soon after a page transition). Read traffic will probably touch
10  * a larger span of pages, but in any case a fairly small number of page
11  * buffers should be sufficient. So, we just search the buffers using plain
12  * linear search; there's no need for a hashtable or anything fancy.
13  * The management algorithm is straight LRU except that we will never swap
14  * out the latest page (since we know it's going to be hit again eventually).
15  *
16  * We use a control LWLock to protect the shared data structures, plus
17  * per-buffer LWLocks that synchronize I/O for each buffer. The control lock
18  * must be held to examine or modify any shared state. A process that is
19  * reading in or writing out a page buffer does not hold the control lock,
20  * only the per-buffer lock for the buffer it is working on.
21  *
22  * "Holding the control lock" means exclusive lock in all cases except for
23  * SimpleLruReadPage_ReadOnly(); see comments for SlruRecentlyUsed() for
24  * the implications of that.
25  *
26  * When initiating I/O on a buffer, we acquire the per-buffer lock exclusively
27  * before releasing the control lock. The per-buffer lock is released after
28  * completing the I/O, re-acquiring the control lock, and updating the shared
29  * state. (Deadlock is not possible here, because we never try to initiate
30  * I/O when someone else is already doing I/O on the same buffer.)
31  * To wait for I/O to complete, release the control lock, acquire the
32  * per-buffer lock in shared mode, immediately release the per-buffer lock,
33  * reacquire the control lock, and then recheck state (since arbitrary things
34  * could have happened while we didn't have the lock).
35  *
36  * As with the regular buffer manager, it is possible for another process
37  * to re-dirty a page that is currently being written out. This is handled
38  * by re-setting the page's page_dirty flag.
39  *
40  *
41  * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
42  * Portions Copyright (c) 1994, Regents of the University of California
43  *
44  * src/backend/access/transam/slru.c
45  *
46  *-------------------------------------------------------------------------
47  */
48 #include "postgres.h"
49 
50 #include <fcntl.h>
51 #include <sys/stat.h>
52 #include <unistd.h>
53 
54 #include "access/slru.h"
55 #include "access/transam.h"
56 #include "access/xlog.h"
57 #include "pgstat.h"
58 #include "storage/fd.h"
59 #include "storage/shmem.h"
60 #include "miscadmin.h"
61 
62 
63 #define SlruFileName(ctl, path, seg) \
64  snprintf(path, MAXPGPATH, "%s/%04X", (ctl)->Dir, seg)
65 
66 /*
67  * During SimpleLruFlush(), we will usually not need to write/fsync more
68  * than one or two physical files, but we may need to write several pages
69  * per file. We can consolidate the I/O requests by leaving files open
70  * until control returns to SimpleLruFlush(). This data structure remembers
71  * which files are open.
72  */
73 #define MAX_FLUSH_BUFFERS 16
74 
75 typedef struct SlruFlushData
76 {
77  int num_files; /* # files actually open */
78  int fd[MAX_FLUSH_BUFFERS]; /* their FD's */
79  int segno[MAX_FLUSH_BUFFERS]; /* their log seg#s */
81 
82 typedef struct SlruFlushData *SlruFlush;
83 
84 /*
85  * Macro to mark a buffer slot "most recently used". Note multiple evaluation
86  * of arguments!
87  *
88  * The reason for the if-test is that there are often many consecutive
89  * accesses to the same page (particularly the latest page). By suppressing
90  * useless increments of cur_lru_count, we reduce the probability that old
91  * pages' counts will "wrap around" and make them appear recently used.
92  *
93  * We allow this code to be executed concurrently by multiple processes within
94  * SimpleLruReadPage_ReadOnly(). As long as int reads and writes are atomic,
95  * this should not cause any completely-bogus values to enter the computation.
96  * However, it is possible for either cur_lru_count or individual
97  * page_lru_count entries to be "reset" to lower values than they should have,
98  * in case a process is delayed while it executes this macro. With care in
99  * SlruSelectLRUPage(), this does little harm, and in any case the absolute
100  * worst possible consequence is a nonoptimal choice of page to evict. The
101  * gain from allowing concurrent reads of SLRU pages seems worth it.
102  */
103 #define SlruRecentlyUsed(shared, slotno) \
104  do { \
105  int new_lru_count = (shared)->cur_lru_count; \
106  if (new_lru_count != (shared)->page_lru_count[slotno]) { \
107  (shared)->cur_lru_count = ++new_lru_count; \
108  (shared)->page_lru_count[slotno] = new_lru_count; \
109  } \
110  } while (0)
111 
112 /* Saved info for SlruReportIOError */
113 typedef enum
114 {
122 
124 static int slru_errno;
125 
126 
127 static void SimpleLruZeroLSNs(SlruCtl ctl, int slotno);
128 static void SimpleLruWaitIO(SlruCtl ctl, int slotno);
129 static void SlruInternalWritePage(SlruCtl ctl, int slotno, SlruFlush fdata);
130 static bool SlruPhysicalReadPage(SlruCtl ctl, int pageno, int slotno);
131 static bool SlruPhysicalWritePage(SlruCtl ctl, int pageno, int slotno,
132  SlruFlush fdata);
133 static void SlruReportIOError(SlruCtl ctl, int pageno, TransactionId xid);
134 static int SlruSelectLRUPage(SlruCtl ctl, int pageno);
135 
136 static bool SlruScanDirCbDeleteCutoff(SlruCtl ctl, char *filename,
137  int segpage, void *data);
138 static void SlruInternalDeleteSegment(SlruCtl ctl, char *filename);
139 
140 /*
141  * Initialization of shared memory
142  */
143 
144 Size
145 SimpleLruShmemSize(int nslots, int nlsns)
146 {
147  Size sz;
148 
149  /* we assume nslots isn't so large as to risk overflow */
150  sz = MAXALIGN(sizeof(SlruSharedData));
151  sz += MAXALIGN(nslots * sizeof(char *)); /* page_buffer[] */
152  sz += MAXALIGN(nslots * sizeof(SlruPageStatus)); /* page_status[] */
153  sz += MAXALIGN(nslots * sizeof(bool)); /* page_dirty[] */
154  sz += MAXALIGN(nslots * sizeof(int)); /* page_number[] */
155  sz += MAXALIGN(nslots * sizeof(int)); /* page_lru_count[] */
156  sz += MAXALIGN(nslots * sizeof(LWLockPadded)); /* buffer_locks[] */
157 
158  if (nlsns > 0)
159  sz += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr)); /* group_lsn[] */
160 
161  return BUFFERALIGN(sz) + BLCKSZ * nslots;
162 }
163 
164 void
165 SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
166  LWLock *ctllock, const char *subdir, int tranche_id)
167 {
168  SlruShared shared;
169  bool found;
170 
171  shared = (SlruShared) ShmemInitStruct(name,
172  SimpleLruShmemSize(nslots, nlsns),
173  &found);
174 
175  if (!IsUnderPostmaster)
176  {
177  /* Initialize locks and shared memory area */
178  char *ptr;
179  Size offset;
180  int slotno;
181 
182  Assert(!found);
183 
184  memset(shared, 0, sizeof(SlruSharedData));
185 
186  shared->ControlLock = ctllock;
187 
188  shared->num_slots = nslots;
189  shared->lsn_groups_per_page = nlsns;
190 
191  shared->cur_lru_count = 0;
192 
193  /* shared->latest_page_number will be set later */
194 
195  ptr = (char *) shared;
196  offset = MAXALIGN(sizeof(SlruSharedData));
197  shared->page_buffer = (char **) (ptr + offset);
198  offset += MAXALIGN(nslots * sizeof(char *));
199  shared->page_status = (SlruPageStatus *) (ptr + offset);
200  offset += MAXALIGN(nslots * sizeof(SlruPageStatus));
201  shared->page_dirty = (bool *) (ptr + offset);
202  offset += MAXALIGN(nslots * sizeof(bool));
203  shared->page_number = (int *) (ptr + offset);
204  offset += MAXALIGN(nslots * sizeof(int));
205  shared->page_lru_count = (int *) (ptr + offset);
206  offset += MAXALIGN(nslots * sizeof(int));
207 
208  /* Initialize LWLocks */
209  shared->buffer_locks = (LWLockPadded *) (ptr + offset);
210  offset += MAXALIGN(nslots * sizeof(LWLockPadded));
211 
212  if (nlsns > 0)
213  {
214  shared->group_lsn = (XLogRecPtr *) (ptr + offset);
215  offset += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr));
216  }
217 
218  Assert(strlen(name) + 1 < SLRU_MAX_NAME_LENGTH);
220  shared->lwlock_tranche_id = tranche_id;
221 
222  ptr += BUFFERALIGN(offset);
223  for (slotno = 0; slotno < nslots; slotno++)
224  {
225  LWLockInitialize(&shared->buffer_locks[slotno].lock,
226  shared->lwlock_tranche_id);
227 
228  shared->page_buffer[slotno] = ptr;
229  shared->page_status[slotno] = SLRU_PAGE_EMPTY;
230  shared->page_dirty[slotno] = false;
231  shared->page_lru_count[slotno] = 0;
232  ptr += BLCKSZ;
233  }
234 
235  /* Should fit to estimated shmem size */
236  Assert(ptr - (char *) shared <= SimpleLruShmemSize(nslots, nlsns));
237  }
238  else
239  Assert(found);
240 
241  /* Register SLRU tranche in the main tranches array */
243  shared->lwlock_tranche_name);
244 
245  /*
246  * Initialize the unshared control struct, including directory path. We
247  * assume caller set PagePrecedes.
248  */
249  ctl->shared = shared;
250  ctl->do_fsync = true; /* default behavior */
251  StrNCpy(ctl->Dir, subdir, sizeof(ctl->Dir));
252 }
253 
254 /*
255  * Initialize (or reinitialize) a page to zeroes.
256  *
257  * The page is not actually written, just set up in shared memory.
258  * The slot number of the new page is returned.
259  *
260  * Control lock must be held at entry, and will be held at exit.
261  */
262 int
263 SimpleLruZeroPage(SlruCtl ctl, int pageno)
264 {
265  SlruShared shared = ctl->shared;
266  int slotno;
267 
268  /* Find a suitable buffer slot for the page */
269  slotno = SlruSelectLRUPage(ctl, pageno);
270  Assert(shared->page_status[slotno] == SLRU_PAGE_EMPTY ||
271  (shared->page_status[slotno] == SLRU_PAGE_VALID &&
272  !shared->page_dirty[slotno]) ||
273  shared->page_number[slotno] == pageno);
274 
275  /* Mark the slot as containing this page */
276  shared->page_number[slotno] = pageno;
277  shared->page_status[slotno] = SLRU_PAGE_VALID;
278  shared->page_dirty[slotno] = true;
279  SlruRecentlyUsed(shared, slotno);
280 
281  /* Set the buffer to zeroes */
282  MemSet(shared->page_buffer[slotno], 0, BLCKSZ);
283 
284  /* Set the LSNs for this new page to zero */
285  SimpleLruZeroLSNs(ctl, slotno);
286 
287  /* Assume this page is now the latest active page */
288  shared->latest_page_number = pageno;
289 
290  return slotno;
291 }
292 
293 /*
294  * Zero all the LSNs we store for this slru page.
295  *
296  * This should be called each time we create a new page, and each time we read
297  * in a page from disk into an existing buffer. (Such an old page cannot
298  * have any interesting LSNs, since we'd have flushed them before writing
299  * the page in the first place.)
300  *
301  * This assumes that InvalidXLogRecPtr is bitwise-all-0.
302  */
303 static void
304 SimpleLruZeroLSNs(SlruCtl ctl, int slotno)
305 {
306  SlruShared shared = ctl->shared;
307 
308  if (shared->lsn_groups_per_page > 0)
309  MemSet(&shared->group_lsn[slotno * shared->lsn_groups_per_page], 0,
310  shared->lsn_groups_per_page * sizeof(XLogRecPtr));
311 }
312 
313 /*
314  * Wait for any active I/O on a page slot to finish. (This does not
315  * guarantee that new I/O hasn't been started before we return, though.
316  * In fact the slot might not even contain the same page anymore.)
317  *
318  * Control lock must be held at entry, and will be held at exit.
319  */
320 static void
321 SimpleLruWaitIO(SlruCtl ctl, int slotno)
322 {
323  SlruShared shared = ctl->shared;
324 
325  /* See notes at top of file */
326  LWLockRelease(shared->ControlLock);
327  LWLockAcquire(&shared->buffer_locks[slotno].lock, LW_SHARED);
328  LWLockRelease(&shared->buffer_locks[slotno].lock);
330 
331  /*
332  * If the slot is still in an io-in-progress state, then either someone
333  * already started a new I/O on the slot, or a previous I/O failed and
334  * neglected to reset the page state. That shouldn't happen, really, but
335  * it seems worth a few extra cycles to check and recover from it. We can
336  * cheaply test for failure by seeing if the buffer lock is still held (we
337  * assume that transaction abort would release the lock).
338  */
339  if (shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS ||
340  shared->page_status[slotno] == SLRU_PAGE_WRITE_IN_PROGRESS)
341  {
342  if (LWLockConditionalAcquire(&shared->buffer_locks[slotno].lock, LW_SHARED))
343  {
344  /* indeed, the I/O must have failed */
345  if (shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS)
346  shared->page_status[slotno] = SLRU_PAGE_EMPTY;
347  else /* write_in_progress */
348  {
349  shared->page_status[slotno] = SLRU_PAGE_VALID;
350  shared->page_dirty[slotno] = true;
351  }
352  LWLockRelease(&shared->buffer_locks[slotno].lock);
353  }
354  }
355 }
356 
357 /*
358  * Find a page in a shared buffer, reading it in if necessary.
359  * The page number must correspond to an already-initialized page.
360  *
361  * If write_ok is true then it is OK to return a page that is in
362  * WRITE_IN_PROGRESS state; it is the caller's responsibility to be sure
363  * that modification of the page is safe. If write_ok is false then we
364  * will not return the page until it is not undergoing active I/O.
365  *
366  * The passed-in xid is used only for error reporting, and may be
367  * InvalidTransactionId if no specific xid is associated with the action.
368  *
369  * Return value is the shared-buffer slot number now holding the page.
370  * The buffer's LRU access info is updated.
371  *
372  * Control lock must be held at entry, and will be held at exit.
373  */
374 int
375 SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok,
376  TransactionId xid)
377 {
378  SlruShared shared = ctl->shared;
379 
380  /* Outer loop handles restart if we must wait for someone else's I/O */
381  for (;;)
382  {
383  int slotno;
384  bool ok;
385 
386  /* See if page already is in memory; if not, pick victim slot */
387  slotno = SlruSelectLRUPage(ctl, pageno);
388 
389  /* Did we find the page in memory? */
390  if (shared->page_number[slotno] == pageno &&
391  shared->page_status[slotno] != SLRU_PAGE_EMPTY)
392  {
393  /*
394  * If page is still being read in, we must wait for I/O. Likewise
395  * if the page is being written and the caller said that's not OK.
396  */
397  if (shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS ||
398  (shared->page_status[slotno] == SLRU_PAGE_WRITE_IN_PROGRESS &&
399  !write_ok))
400  {
401  SimpleLruWaitIO(ctl, slotno);
402  /* Now we must recheck state from the top */
403  continue;
404  }
405  /* Otherwise, it's ready to use */
406  SlruRecentlyUsed(shared, slotno);
407  return slotno;
408  }
409 
410  /* We found no match; assert we selected a freeable slot */
411  Assert(shared->page_status[slotno] == SLRU_PAGE_EMPTY ||
412  (shared->page_status[slotno] == SLRU_PAGE_VALID &&
413  !shared->page_dirty[slotno]));
414 
415  /* Mark the slot read-busy */
416  shared->page_number[slotno] = pageno;
417  shared->page_status[slotno] = SLRU_PAGE_READ_IN_PROGRESS;
418  shared->page_dirty[slotno] = false;
419 
420  /* Acquire per-buffer lock (cannot deadlock, see notes at top) */
421  LWLockAcquire(&shared->buffer_locks[slotno].lock, LW_EXCLUSIVE);
422 
423  /* Release control lock while doing I/O */
424  LWLockRelease(shared->ControlLock);
425 
426  /* Do the read */
427  ok = SlruPhysicalReadPage(ctl, pageno, slotno);
428 
429  /* Set the LSNs for this newly read-in page to zero */
430  SimpleLruZeroLSNs(ctl, slotno);
431 
432  /* Re-acquire control lock and update page state */
434 
435  Assert(shared->page_number[slotno] == pageno &&
436  shared->page_status[slotno] == SLRU_PAGE_READ_IN_PROGRESS &&
437  !shared->page_dirty[slotno]);
438 
439  shared->page_status[slotno] = ok ? SLRU_PAGE_VALID : SLRU_PAGE_EMPTY;
440 
441  LWLockRelease(&shared->buffer_locks[slotno].lock);
442 
443  /* Now it's okay to ereport if we failed */
444  if (!ok)
445  SlruReportIOError(ctl, pageno, xid);
446 
447  SlruRecentlyUsed(shared, slotno);
448  return slotno;
449  }
450 }
451 
452 /*
453  * Find a page in a shared buffer, reading it in if necessary.
454  * The page number must correspond to an already-initialized page.
455  * The caller must intend only read-only access to the page.
456  *
457  * The passed-in xid is used only for error reporting, and may be
458  * InvalidTransactionId if no specific xid is associated with the action.
459  *
460  * Return value is the shared-buffer slot number now holding the page.
461  * The buffer's LRU access info is updated.
462  *
463  * Control lock must NOT be held at entry, but will be held at exit.
464  * It is unspecified whether the lock will be shared or exclusive.
465  */
466 int
468 {
469  SlruShared shared = ctl->shared;
470  int slotno;
471 
472  /* Try to find the page while holding only shared lock */
474 
475  /* See if page is already in a buffer */
476  for (slotno = 0; slotno < shared->num_slots; slotno++)
477  {
478  if (shared->page_number[slotno] == pageno &&
479  shared->page_status[slotno] != SLRU_PAGE_EMPTY &&
480  shared->page_status[slotno] != SLRU_PAGE_READ_IN_PROGRESS)
481  {
482  /* See comments for SlruRecentlyUsed macro */
483  SlruRecentlyUsed(shared, slotno);
484  return slotno;
485  }
486  }
487 
488  /* No luck, so switch to normal exclusive lock and do regular read */
489  LWLockRelease(shared->ControlLock);
491 
492  return SimpleLruReadPage(ctl, pageno, true, xid);
493 }
494 
495 /*
496  * Write a page from a shared buffer, if necessary.
497  * Does nothing if the specified slot is not dirty.
498  *
499  * NOTE: only one write attempt is made here. Hence, it is possible that
500  * the page is still dirty at exit (if someone else re-dirtied it during
501  * the write). However, we *do* attempt a fresh write even if the page
502  * is already being written; this is for checkpoints.
503  *
504  * Control lock must be held at entry, and will be held at exit.
505  */
506 static void
507 SlruInternalWritePage(SlruCtl ctl, int slotno, SlruFlush fdata)
508 {
509  SlruShared shared = ctl->shared;
510  int pageno = shared->page_number[slotno];
511  bool ok;
512 
513  /* If a write is in progress, wait for it to finish */
514  while (shared->page_status[slotno] == SLRU_PAGE_WRITE_IN_PROGRESS &&
515  shared->page_number[slotno] == pageno)
516  {
517  SimpleLruWaitIO(ctl, slotno);
518  }
519 
520  /*
521  * Do nothing if page is not dirty, or if buffer no longer contains the
522  * same page we were called for.
523  */
524  if (!shared->page_dirty[slotno] ||
525  shared->page_status[slotno] != SLRU_PAGE_VALID ||
526  shared->page_number[slotno] != pageno)
527  return;
528 
529  /*
530  * Mark the slot write-busy, and clear the dirtybit. After this point, a
531  * transaction status update on this page will mark it dirty again.
532  */
533  shared->page_status[slotno] = SLRU_PAGE_WRITE_IN_PROGRESS;
534  shared->page_dirty[slotno] = false;
535 
536  /* Acquire per-buffer lock (cannot deadlock, see notes at top) */
537  LWLockAcquire(&shared->buffer_locks[slotno].lock, LW_EXCLUSIVE);
538 
539  /* Release control lock while doing I/O */
540  LWLockRelease(shared->ControlLock);
541 
542  /* Do the write */
543  ok = SlruPhysicalWritePage(ctl, pageno, slotno, fdata);
544 
545  /* If we failed, and we're in a flush, better close the files */
546  if (!ok && fdata)
547  {
548  int i;
549 
550  for (i = 0; i < fdata->num_files; i++)
551  CloseTransientFile(fdata->fd[i]);
552  }
553 
554  /* Re-acquire control lock and update page state */
556 
557  Assert(shared->page_number[slotno] == pageno &&
558  shared->page_status[slotno] == SLRU_PAGE_WRITE_IN_PROGRESS);
559 
560  /* If we failed to write, mark the page dirty again */
561  if (!ok)
562  shared->page_dirty[slotno] = true;
563 
564  shared->page_status[slotno] = SLRU_PAGE_VALID;
565 
566  LWLockRelease(&shared->buffer_locks[slotno].lock);
567 
568  /* Now it's okay to ereport if we failed */
569  if (!ok)
571 }
572 
573 /*
574  * Wrapper of SlruInternalWritePage, for external callers.
575  * fdata is always passed a NULL here.
576  */
577 void
578 SimpleLruWritePage(SlruCtl ctl, int slotno)
579 {
580  SlruInternalWritePage(ctl, slotno, NULL);
581 }
582 
583 /*
584  * Return whether the given page exists on disk.
585  *
586  * A false return means that either the file does not exist, or that it's not
587  * large enough to contain the given page.
588  */
589 bool
591 {
592  int segno = pageno / SLRU_PAGES_PER_SEGMENT;
593  int rpageno = pageno % SLRU_PAGES_PER_SEGMENT;
594  int offset = rpageno * BLCKSZ;
595  char path[MAXPGPATH];
596  int fd;
597  bool result;
598  off_t endpos;
599 
600  SlruFileName(ctl, path, segno);
601 
602  fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
603  if (fd < 0)
604  {
605  /* expected: file doesn't exist */
606  if (errno == ENOENT)
607  return false;
608 
609  /* report error normally */
611  slru_errno = errno;
612  SlruReportIOError(ctl, pageno, 0);
613  }
614 
615  if ((endpos = lseek(fd, 0, SEEK_END)) < 0)
616  {
618  slru_errno = errno;
619  SlruReportIOError(ctl, pageno, 0);
620  }
621 
622  result = endpos >= (off_t) (offset + BLCKSZ);
623 
624  CloseTransientFile(fd);
625  return result;
626 }
627 
628 /*
629  * Physical read of a (previously existing) page into a buffer slot
630  *
631  * On failure, we cannot just ereport(ERROR) since caller has put state in
632  * shared memory that must be undone. So, we return false and save enough
633  * info in static variables to let SlruReportIOError make the report.
634  *
635  * For now, assume it's not worth keeping a file pointer open across
636  * read/write operations. We could cache one virtual file pointer ...
637  */
638 static bool
639 SlruPhysicalReadPage(SlruCtl ctl, int pageno, int slotno)
640 {
641  SlruShared shared = ctl->shared;
642  int segno = pageno / SLRU_PAGES_PER_SEGMENT;
643  int rpageno = pageno % SLRU_PAGES_PER_SEGMENT;
644  int offset = rpageno * BLCKSZ;
645  char path[MAXPGPATH];
646  int fd;
647 
648  SlruFileName(ctl, path, segno);
649 
650  /*
651  * In a crash-and-restart situation, it's possible for us to receive
652  * commands to set the commit status of transactions whose bits are in
653  * already-truncated segments of the commit log (see notes in
654  * SlruPhysicalWritePage). Hence, if we are InRecovery, allow the case
655  * where the file doesn't exist, and return zeroes instead.
656  */
657  fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
658  if (fd < 0)
659  {
660  if (errno != ENOENT || !InRecovery)
661  {
663  slru_errno = errno;
664  return false;
665  }
666 
667  ereport(LOG,
668  (errmsg("file \"%s\" doesn't exist, reading as zeroes",
669  path)));
670  MemSet(shared->page_buffer[slotno], 0, BLCKSZ);
671  return true;
672  }
673 
674  if (lseek(fd, (off_t) offset, SEEK_SET) < 0)
675  {
677  slru_errno = errno;
678  CloseTransientFile(fd);
679  return false;
680  }
681 
682  errno = 0;
684  if (read(fd, shared->page_buffer[slotno], BLCKSZ) != BLCKSZ)
685  {
688  slru_errno = errno;
689  CloseTransientFile(fd);
690  return false;
691  }
693 
694  if (CloseTransientFile(fd))
695  {
697  slru_errno = errno;
698  return false;
699  }
700 
701  return true;
702 }
703 
704 /*
705  * Physical write of a page from a buffer slot
706  *
707  * On failure, we cannot just ereport(ERROR) since caller has put state in
708  * shared memory that must be undone. So, we return false and save enough
709  * info in static variables to let SlruReportIOError make the report.
710  *
711  * For now, assume it's not worth keeping a file pointer open across
712  * independent read/write operations. We do batch operations during
713  * SimpleLruFlush, though.
714  *
715  * fdata is NULL for a standalone write, pointer to open-file info during
716  * SimpleLruFlush.
717  */
718 static bool
719 SlruPhysicalWritePage(SlruCtl ctl, int pageno, int slotno, SlruFlush fdata)
720 {
721  SlruShared shared = ctl->shared;
722  int segno = pageno / SLRU_PAGES_PER_SEGMENT;
723  int rpageno = pageno % SLRU_PAGES_PER_SEGMENT;
724  int offset = rpageno * BLCKSZ;
725  char path[MAXPGPATH];
726  int fd = -1;
727 
728  /*
729  * Honor the write-WAL-before-data rule, if appropriate, so that we do not
730  * write out data before associated WAL records. This is the same action
731  * performed during FlushBuffer() in the main buffer manager.
732  */
733  if (shared->group_lsn != NULL)
734  {
735  /*
736  * We must determine the largest async-commit LSN for the page. This
737  * is a bit tedious, but since this entire function is a slow path
738  * anyway, it seems better to do this here than to maintain a per-page
739  * LSN variable (which'd need an extra comparison in the
740  * transaction-commit path).
741  */
742  XLogRecPtr max_lsn;
743  int lsnindex,
744  lsnoff;
745 
746  lsnindex = slotno * shared->lsn_groups_per_page;
747  max_lsn = shared->group_lsn[lsnindex++];
748  for (lsnoff = 1; lsnoff < shared->lsn_groups_per_page; lsnoff++)
749  {
750  XLogRecPtr this_lsn = shared->group_lsn[lsnindex++];
751 
752  if (max_lsn < this_lsn)
753  max_lsn = this_lsn;
754  }
755 
756  if (!XLogRecPtrIsInvalid(max_lsn))
757  {
758  /*
759  * As noted above, elog(ERROR) is not acceptable here, so if
760  * XLogFlush were to fail, we must PANIC. This isn't much of a
761  * restriction because XLogFlush is just about all critical
762  * section anyway, but let's make sure.
763  */
765  XLogFlush(max_lsn);
767  }
768  }
769 
770  /*
771  * During a Flush, we may already have the desired file open.
772  */
773  if (fdata)
774  {
775  int i;
776 
777  for (i = 0; i < fdata->num_files; i++)
778  {
779  if (fdata->segno[i] == segno)
780  {
781  fd = fdata->fd[i];
782  break;
783  }
784  }
785  }
786 
787  if (fd < 0)
788  {
789  /*
790  * If the file doesn't already exist, we should create it. It is
791  * possible for this to need to happen when writing a page that's not
792  * first in its segment; we assume the OS can cope with that. (Note:
793  * it might seem that it'd be okay to create files only when
794  * SimpleLruZeroPage is called for the first page of a segment.
795  * However, if after a crash and restart the REDO logic elects to
796  * replay the log from a checkpoint before the latest one, then it's
797  * possible that we will get commands to set transaction status of
798  * transactions that have already been truncated from the commit log.
799  * Easiest way to deal with that is to accept references to
800  * nonexistent files here and in SlruPhysicalReadPage.)
801  *
802  * Note: it is possible for more than one backend to be executing this
803  * code simultaneously for different pages of the same file. Hence,
804  * don't use O_EXCL or O_TRUNC or anything like that.
805  */
806  SlruFileName(ctl, path, segno);
807  fd = OpenTransientFile(path, O_RDWR | O_CREAT | PG_BINARY);
808  if (fd < 0)
809  {
811  slru_errno = errno;
812  return false;
813  }
814 
815  if (fdata)
816  {
817  if (fdata->num_files < MAX_FLUSH_BUFFERS)
818  {
819  fdata->fd[fdata->num_files] = fd;
820  fdata->segno[fdata->num_files] = segno;
821  fdata->num_files++;
822  }
823  else
824  {
825  /*
826  * In the unlikely event that we exceed MAX_FLUSH_BUFFERS,
827  * fall back to treating it as a standalone write.
828  */
829  fdata = NULL;
830  }
831  }
832  }
833 
834  if (lseek(fd, (off_t) offset, SEEK_SET) < 0)
835  {
837  slru_errno = errno;
838  if (!fdata)
839  CloseTransientFile(fd);
840  return false;
841  }
842 
843  errno = 0;
845  if (write(fd, shared->page_buffer[slotno], BLCKSZ) != BLCKSZ)
846  {
848  /* if write didn't set errno, assume problem is no disk space */
849  if (errno == 0)
850  errno = ENOSPC;
852  slru_errno = errno;
853  if (!fdata)
854  CloseTransientFile(fd);
855  return false;
856  }
858 
859  /*
860  * If not part of Flush, need to fsync now. We assume this happens
861  * infrequently enough that it's not a performance issue.
862  */
863  if (!fdata)
864  {
866  if (ctl->do_fsync && pg_fsync(fd))
867  {
870  slru_errno = errno;
871  CloseTransientFile(fd);
872  return false;
873  }
875 
876  if (CloseTransientFile(fd))
877  {
879  slru_errno = errno;
880  return false;
881  }
882  }
883 
884  return true;
885 }
886 
887 /*
888  * Issue the error message after failure of SlruPhysicalReadPage or
889  * SlruPhysicalWritePage. Call this after cleaning up shared-memory state.
890  */
891 static void
893 {
894  int segno = pageno / SLRU_PAGES_PER_SEGMENT;
895  int rpageno = pageno % SLRU_PAGES_PER_SEGMENT;
896  int offset = rpageno * BLCKSZ;
897  char path[MAXPGPATH];
898 
899  SlruFileName(ctl, path, segno);
900  errno = slru_errno;
901  switch (slru_errcause)
902  {
903  case SLRU_OPEN_FAILED:
904  ereport(ERROR,
906  errmsg("could not access status of transaction %u", xid),
907  errdetail("Could not open file \"%s\": %m.", path)));
908  break;
909  case SLRU_SEEK_FAILED:
910  ereport(ERROR,
912  errmsg("could not access status of transaction %u", xid),
913  errdetail("Could not seek in file \"%s\" to offset %u: %m.",
914  path, offset)));
915  break;
916  case SLRU_READ_FAILED:
917  ereport(ERROR,
919  errmsg("could not access status of transaction %u", xid),
920  errdetail("Could not read from file \"%s\" at offset %u: %m.",
921  path, offset)));
922  break;
923  case SLRU_WRITE_FAILED:
924  ereport(ERROR,
926  errmsg("could not access status of transaction %u", xid),
927  errdetail("Could not write to file \"%s\" at offset %u: %m.",
928  path, offset)));
929  break;
930  case SLRU_FSYNC_FAILED:
931  ereport(ERROR,
933  errmsg("could not access status of transaction %u", xid),
934  errdetail("Could not fsync file \"%s\": %m.",
935  path)));
936  break;
937  case SLRU_CLOSE_FAILED:
938  ereport(ERROR,
940  errmsg("could not access status of transaction %u", xid),
941  errdetail("Could not close file \"%s\": %m.",
942  path)));
943  break;
944  default:
945  /* can't get here, we trust */
946  elog(ERROR, "unrecognized SimpleLru error cause: %d",
947  (int) slru_errcause);
948  break;
949  }
950 }
951 
952 /*
953  * Select the slot to re-use when we need a free slot.
954  *
955  * The target page number is passed because we need to consider the
956  * possibility that some other process reads in the target page while
957  * we are doing I/O to free a slot. Hence, check or recheck to see if
958  * any slot already holds the target page, and return that slot if so.
959  * Thus, the returned slot is *either* a slot already holding the pageno
960  * (could be any state except EMPTY), *or* a freeable slot (state EMPTY
961  * or CLEAN).
962  *
963  * Control lock must be held at entry, and will be held at exit.
964  */
965 static int
966 SlruSelectLRUPage(SlruCtl ctl, int pageno)
967 {
968  SlruShared shared = ctl->shared;
969 
970  /* Outer loop handles restart after I/O */
971  for (;;)
972  {
973  int slotno;
974  int cur_count;
975  int bestvalidslot = 0; /* keep compiler quiet */
976  int best_valid_delta = -1;
977  int best_valid_page_number = 0; /* keep compiler quiet */
978  int bestinvalidslot = 0; /* keep compiler quiet */
979  int best_invalid_delta = -1;
980  int best_invalid_page_number = 0; /* keep compiler quiet */
981 
982  /* See if page already has a buffer assigned */
983  for (slotno = 0; slotno < shared->num_slots; slotno++)
984  {
985  if (shared->page_number[slotno] == pageno &&
986  shared->page_status[slotno] != SLRU_PAGE_EMPTY)
987  return slotno;
988  }
989 
990  /*
991  * If we find any EMPTY slot, just select that one. Else choose a
992  * victim page to replace. We normally take the least recently used
993  * valid page, but we will never take the slot containing
994  * latest_page_number, even if it appears least recently used. We
995  * will select a slot that is already I/O busy only if there is no
996  * other choice: a read-busy slot will not be least recently used once
997  * the read finishes, and waiting for an I/O on a write-busy slot is
998  * inferior to just picking some other slot. Testing shows the slot
999  * we pick instead will often be clean, allowing us to begin a read at
1000  * once.
1001  *
1002  * Normally the page_lru_count values will all be different and so
1003  * there will be a well-defined LRU page. But since we allow
1004  * concurrent execution of SlruRecentlyUsed() within
1005  * SimpleLruReadPage_ReadOnly(), it is possible that multiple pages
1006  * acquire the same lru_count values. In that case we break ties by
1007  * choosing the furthest-back page.
1008  *
1009  * Notice that this next line forcibly advances cur_lru_count to a
1010  * value that is certainly beyond any value that will be in the
1011  * page_lru_count array after the loop finishes. This ensures that
1012  * the next execution of SlruRecentlyUsed will mark the page newly
1013  * used, even if it's for a page that has the current counter value.
1014  * That gets us back on the path to having good data when there are
1015  * multiple pages with the same lru_count.
1016  */
1017  cur_count = (shared->cur_lru_count)++;
1018  for (slotno = 0; slotno < shared->num_slots; slotno++)
1019  {
1020  int this_delta;
1021  int this_page_number;
1022 
1023  if (shared->page_status[slotno] == SLRU_PAGE_EMPTY)
1024  return slotno;
1025  this_delta = cur_count - shared->page_lru_count[slotno];
1026  if (this_delta < 0)
1027  {
1028  /*
1029  * Clean up in case shared updates have caused cur_count
1030  * increments to get "lost". We back off the page counts,
1031  * rather than trying to increase cur_count, to avoid any
1032  * question of infinite loops or failure in the presence of
1033  * wrapped-around counts.
1034  */
1035  shared->page_lru_count[slotno] = cur_count;
1036  this_delta = 0;
1037  }
1038  this_page_number = shared->page_number[slotno];
1039  if (this_page_number == shared->latest_page_number)
1040  continue;
1041  if (shared->page_status[slotno] == SLRU_PAGE_VALID)
1042  {
1043  if (this_delta > best_valid_delta ||
1044  (this_delta == best_valid_delta &&
1045  ctl->PagePrecedes(this_page_number,
1046  best_valid_page_number)))
1047  {
1048  bestvalidslot = slotno;
1049  best_valid_delta = this_delta;
1050  best_valid_page_number = this_page_number;
1051  }
1052  }
1053  else
1054  {
1055  if (this_delta > best_invalid_delta ||
1056  (this_delta == best_invalid_delta &&
1057  ctl->PagePrecedes(this_page_number,
1058  best_invalid_page_number)))
1059  {
1060  bestinvalidslot = slotno;
1061  best_invalid_delta = this_delta;
1062  best_invalid_page_number = this_page_number;
1063  }
1064  }
1065  }
1066 
1067  /*
1068  * If all pages (except possibly the latest one) are I/O busy, we'll
1069  * have to wait for an I/O to complete and then retry. In that
1070  * unhappy case, we choose to wait for the I/O on the least recently
1071  * used slot, on the assumption that it was likely initiated first of
1072  * all the I/Os in progress and may therefore finish first.
1073  */
1074  if (best_valid_delta < 0)
1075  {
1076  SimpleLruWaitIO(ctl, bestinvalidslot);
1077  continue;
1078  }
1079 
1080  /*
1081  * If the selected page is clean, we're set.
1082  */
1083  if (!shared->page_dirty[bestvalidslot])
1084  return bestvalidslot;
1085 
1086  /*
1087  * Write the page.
1088  */
1089  SlruInternalWritePage(ctl, bestvalidslot, NULL);
1090 
1091  /*
1092  * Now loop back and try again. This is the easiest way of dealing
1093  * with corner cases such as the victim page being re-dirtied while we
1094  * wrote it.
1095  */
1096  }
1097 }
1098 
1099 /*
1100  * Flush dirty pages to disk during checkpoint or database shutdown
1101  */
1102 void
1103 SimpleLruFlush(SlruCtl ctl, bool allow_redirtied)
1104 {
1105  SlruShared shared = ctl->shared;
1106  SlruFlushData fdata;
1107  int slotno;
1108  int pageno = 0;
1109  int i;
1110  bool ok;
1111 
1112  /*
1113  * Find and write dirty pages
1114  */
1115  fdata.num_files = 0;
1116 
1118 
1119  for (slotno = 0; slotno < shared->num_slots; slotno++)
1120  {
1121  SlruInternalWritePage(ctl, slotno, &fdata);
1122 
1123  /*
1124  * In some places (e.g. checkpoints), we cannot assert that the slot
1125  * is clean now, since another process might have re-dirtied it
1126  * already. That's okay.
1127  */
1128  Assert(allow_redirtied ||
1129  shared->page_status[slotno] == SLRU_PAGE_EMPTY ||
1130  (shared->page_status[slotno] == SLRU_PAGE_VALID &&
1131  !shared->page_dirty[slotno]));
1132  }
1133 
1134  LWLockRelease(shared->ControlLock);
1135 
1136  /*
1137  * Now fsync and close any files that were open
1138  */
1139  ok = true;
1140  for (i = 0; i < fdata.num_files; i++)
1141  {
1143  if (ctl->do_fsync && pg_fsync(fdata.fd[i]))
1144  {
1146  slru_errno = errno;
1147  pageno = fdata.segno[i] * SLRU_PAGES_PER_SEGMENT;
1148  ok = false;
1149  }
1151 
1152  if (CloseTransientFile(fdata.fd[i]))
1153  {
1155  slru_errno = errno;
1156  pageno = fdata.segno[i] * SLRU_PAGES_PER_SEGMENT;
1157  ok = false;
1158  }
1159  }
1160  if (!ok)
1162 }
1163 
1164 /*
1165  * Remove all segments before the one holding the passed page number
1166  */
1167 void
1168 SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
1169 {
1170  SlruShared shared = ctl->shared;
1171  int slotno;
1172 
1173  /*
1174  * The cutoff point is the start of the segment containing cutoffPage.
1175  */
1176  cutoffPage -= cutoffPage % SLRU_PAGES_PER_SEGMENT;
1177 
1178  /*
1179  * Scan shared memory and remove any pages preceding the cutoff page, to
1180  * ensure we won't rewrite them later. (Since this is normally called in
1181  * or just after a checkpoint, any dirty pages should have been flushed
1182  * already ... we're just being extra careful here.)
1183  */
1185 
1186 restart:;
1187 
1188  /*
1189  * While we are holding the lock, make an important safety check: the
1190  * planned cutoff point must be <= the current endpoint page. Otherwise we
1191  * have already wrapped around, and proceeding with the truncation would
1192  * risk removing the current segment.
1193  */
1194  if (ctl->PagePrecedes(shared->latest_page_number, cutoffPage))
1195  {
1196  LWLockRelease(shared->ControlLock);
1197  ereport(LOG,
1198  (errmsg("could not truncate directory \"%s\": apparent wraparound",
1199  ctl->Dir)));
1200  return;
1201  }
1202 
1203  for (slotno = 0; slotno < shared->num_slots; slotno++)
1204  {
1205  if (shared->page_status[slotno] == SLRU_PAGE_EMPTY)
1206  continue;
1207  if (!ctl->PagePrecedes(shared->page_number[slotno], cutoffPage))
1208  continue;
1209 
1210  /*
1211  * If page is clean, just change state to EMPTY (expected case).
1212  */
1213  if (shared->page_status[slotno] == SLRU_PAGE_VALID &&
1214  !shared->page_dirty[slotno])
1215  {
1216  shared->page_status[slotno] = SLRU_PAGE_EMPTY;
1217  continue;
1218  }
1219 
1220  /*
1221  * Hmm, we have (or may have) I/O operations acting on the page, so
1222  * we've got to wait for them to finish and then start again. This is
1223  * the same logic as in SlruSelectLRUPage. (XXX if page is dirty,
1224  * wouldn't it be OK to just discard it without writing it? For now,
1225  * keep the logic the same as it was.)
1226  */
1227  if (shared->page_status[slotno] == SLRU_PAGE_VALID)
1228  SlruInternalWritePage(ctl, slotno, NULL);
1229  else
1230  SimpleLruWaitIO(ctl, slotno);
1231  goto restart;
1232  }
1233 
1234  LWLockRelease(shared->ControlLock);
1235 
1236  /* Now we can remove the old segment(s) */
1237  (void) SlruScanDirectory(ctl, SlruScanDirCbDeleteCutoff, &cutoffPage);
1238 }
1239 
1240 /*
1241  * Delete an individual SLRU segment, identified by the filename.
1242  *
1243  * NB: This does not touch the SLRU buffers themselves, callers have to ensure
1244  * they either can't yet contain anything, or have already been cleaned out.
1245  */
1246 static void
1248 {
1249  char path[MAXPGPATH];
1250 
1251  snprintf(path, MAXPGPATH, "%s/%s", ctl->Dir, filename);
1252  ereport(DEBUG2,
1253  (errmsg("removing file \"%s\"", path)));
1254  unlink(path);
1255 }
1256 
1257 /*
1258  * Delete an individual SLRU segment, identified by the segment number.
1259  */
1260 void
1262 {
1263  SlruShared shared = ctl->shared;
1264  int slotno;
1265  char path[MAXPGPATH];
1266  bool did_write;
1267 
1268  /* Clean out any possibly existing references to the segment. */
1270 restart:
1271  did_write = false;
1272  for (slotno = 0; slotno < shared->num_slots; slotno++)
1273  {
1274  int pagesegno = shared->page_number[slotno] / SLRU_PAGES_PER_SEGMENT;
1275 
1276  if (shared->page_status[slotno] == SLRU_PAGE_EMPTY)
1277  continue;
1278 
1279  /* not the segment we're looking for */
1280  if (pagesegno != segno)
1281  continue;
1282 
1283  /* If page is clean, just change state to EMPTY (expected case). */
1284  if (shared->page_status[slotno] == SLRU_PAGE_VALID &&
1285  !shared->page_dirty[slotno])
1286  {
1287  shared->page_status[slotno] = SLRU_PAGE_EMPTY;
1288  continue;
1289  }
1290 
1291  /* Same logic as SimpleLruTruncate() */
1292  if (shared->page_status[slotno] == SLRU_PAGE_VALID)
1293  SlruInternalWritePage(ctl, slotno, NULL);
1294  else
1295  SimpleLruWaitIO(ctl, slotno);
1296 
1297  did_write = true;
1298  }
1299 
1300  /*
1301  * Be extra careful and re-check. The IO functions release the control
1302  * lock, so new pages could have been read in.
1303  */
1304  if (did_write)
1305  goto restart;
1306 
1307  snprintf(path, MAXPGPATH, "%s/%04X", ctl->Dir, segno);
1308  ereport(DEBUG2,
1309  (errmsg("removing file \"%s\"", path)));
1310  unlink(path);
1311 
1312  LWLockRelease(shared->ControlLock);
1313 }
1314 
1315 /*
1316  * SlruScanDirectory callback
1317  * This callback reports true if there's any segment prior to the one
1318  * containing the page passed as "data".
1319  */
1320 bool
1321 SlruScanDirCbReportPresence(SlruCtl ctl, char *filename, int segpage, void *data)
1322 {
1323  int cutoffPage = *(int *) data;
1324 
1325  cutoffPage -= cutoffPage % SLRU_PAGES_PER_SEGMENT;
1326 
1327  if (ctl->PagePrecedes(segpage, cutoffPage))
1328  return true; /* found one; don't iterate any more */
1329 
1330  return false; /* keep going */
1331 }
1332 
1333 /*
1334  * SlruScanDirectory callback.
1335  * This callback deletes segments prior to the one passed in as "data".
1336  */
1337 static bool
1338 SlruScanDirCbDeleteCutoff(SlruCtl ctl, char *filename, int segpage, void *data)
1339 {
1340  int cutoffPage = *(int *) data;
1341 
1342  if (ctl->PagePrecedes(segpage, cutoffPage))
1343  SlruInternalDeleteSegment(ctl, filename);
1344 
1345  return false; /* keep going */
1346 }
1347 
1348 /*
1349  * SlruScanDirectory callback.
1350  * This callback deletes all segments.
1351  */
1352 bool
1353 SlruScanDirCbDeleteAll(SlruCtl ctl, char *filename, int segpage, void *data)
1354 {
1355  SlruInternalDeleteSegment(ctl, filename);
1356 
1357  return false; /* keep going */
1358 }
1359 
1360 /*
1361  * Scan the SimpleLRU directory and apply a callback to each file found in it.
1362  *
1363  * If the callback returns true, the scan is stopped. The last return value
1364  * from the callback is returned.
1365  *
1366  * The callback receives the following arguments: 1. the SlruCtl struct for the
1367  * slru being truncated; 2. the filename being considered; 3. the page number
1368  * for the first page of that file; 4. a pointer to the opaque data given to us
1369  * by the caller.
1370  *
1371  * Note that the ordering in which the directory is scanned is not guaranteed.
1372  *
1373  * Note that no locking is applied.
1374  */
1375 bool
1377 {
1378  bool retval = false;
1379  DIR *cldir;
1380  struct dirent *clde;
1381  int segno;
1382  int segpage;
1383 
1384  cldir = AllocateDir(ctl->Dir);
1385  while ((clde = ReadDir(cldir, ctl->Dir)) != NULL)
1386  {
1387  size_t len;
1388 
1389  len = strlen(clde->d_name);
1390 
1391  if ((len == 4 || len == 5 || len == 6) &&
1392  strspn(clde->d_name, "0123456789ABCDEF") == len)
1393  {
1394  segno = (int) strtol(clde->d_name, NULL, 16);
1395  segpage = segno * SLRU_PAGES_PER_SEGMENT;
1396 
1397  elog(DEBUG2, "SlruScanDirectory invoking callback on %s/%s",
1398  ctl->Dir, clde->d_name);
1399  retval = callback(ctl, clde->d_name, segpage, data);
1400  if (retval)
1401  break;
1402  }
1403  }
1404  FreeDir(cldir);
1405 
1406  return retval;
1407 }
LWLock * ControlLock
Definition: slru.h:57
int * page_number
Definition: slru.h:69
Definition: lwlock.h:32
SlruPageStatus
Definition: slru.h:44
uint32 TransactionId
Definition: c.h:445
static void SlruInternalWritePage(SlruCtl ctl, int slotno, SlruFlush fdata)
Definition: slru.c:507
bool SlruScanDirCbDeleteAll(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.c:1353
#define write(a, b, c)
Definition: win32.h:14
int latest_page_number
Definition: slru.h:99
char ** page_buffer
Definition: slru.h:66
void SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
Definition: slru.c:1168
bool InRecovery
Definition: xlog.c:194
#define END_CRIT_SECTION()
Definition: miscadmin.h:133
SlruErrorCause
Definition: slru.c:113
static bool SlruPhysicalWritePage(SlruCtl ctl, int pageno, int slotno, SlruFlush fdata)
Definition: slru.c:719
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
#define MemSet(start, val, len)
Definition: c.h:853
static SlruErrorCause slru_errcause
Definition: slru.c:123
int cur_lru_count
Definition: slru.h:92
static void SimpleLruZeroLSNs(SlruCtl ctl, int slotno)
Definition: slru.c:304
int lsn_groups_per_page
Definition: slru.h:81
int segno[MAX_FLUSH_BUFFERS]
Definition: slru.c:79
int snprintf(char *str, size_t count, const char *fmt,...) pg_attribute_printf(3
#define LOG
Definition: elog.h:26
Definition: dirent.h:9
Size SimpleLruShmemSize(int nslots, int nlsns)
Definition: slru.c:145
void SimpleLruFlush(SlruCtl ctl, bool allow_redirtied)
Definition: slru.c:1103
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2763
static void SlruReportIOError(SlruCtl ctl, int pageno, TransactionId xid)
Definition: slru.c:892
#define PG_BINARY
Definition: c.h:1025
struct SlruFlushData * SlruFlush
Definition: slru.c:82
void LWLockRegisterTranche(int tranche_id, const char *tranche_name)
Definition: lwlock.c:598
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1721
static void SlruInternalDeleteSegment(SlruCtl ctl, char *filename)
Definition: slru.c:1247
SlruPageStatus * page_status
Definition: slru.h:67
Definition: dirent.c:25
#define ERROR
Definition: elog.h:43
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2173
char lwlock_tranche_name[SLRU_MAX_NAME_LENGTH]
Definition: slru.h:103
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:372
int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok, TransactionId xid)
Definition: slru.c:375
#define MAXPGPATH
static void callback(struct sockaddr *addr, struct sockaddr *mask, void *unused)
Definition: test_ifaddrs.c:48
static XLogRecPtr endpos
Definition: pg_receivewal.c:46
#define DEBUG2
Definition: elog.h:24
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1289
#define MAX_FLUSH_BUFFERS
Definition: slru.c:73
bool IsUnderPostmaster
Definition: globals.c:101
LWLockPadded * buffer_locks
Definition: slru.h:104
int errdetail(const char *fmt,...)
Definition: elog.c:873
int errcode_for_file_access(void)
Definition: elog.c:598
#define InvalidTransactionId
Definition: transam.h:31
XLogRecPtr * group_lsn
Definition: slru.h:80
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2373
bool SimpleLruDoesPhysicalPageExist(SlruCtl ctl, int pageno)
Definition: slru.c:590
void SimpleLruWritePage(SlruCtl ctl, int slotno)
Definition: slru.c:578
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1244
#define ereport(elevel, rest)
Definition: elog.h:122
static void SimpleLruWaitIO(SlruCtl ctl, int slotno)
Definition: slru.c:321
bool do_fsync
Definition: slru.h:121
void LWLockInitialize(LWLock *lock, int tranche_id)
Definition: lwlock.c:673
int CloseTransientFile(int fd)
Definition: fd.c:2343
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
bool(* SlruScanCallback)(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.h:153
SlruSharedData * SlruShared
Definition: slru.h:107
bool SlruScanDirCbReportPresence(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.c:1321
static bool SlruScanDirCbDeleteCutoff(SlruCtl ctl, char *filename, int segpage, void *data)
Definition: slru.c:1338
#define SlruFileName(ctl, path, seg)
Definition: slru.c:63
char Dir[64]
Definition: slru.h:134
int SimpleLruReadPage_ReadOnly(SlruCtl ctl, int pageno, TransactionId xid)
Definition: slru.c:467
#define SLRU_MAX_NAME_LENGTH
Definition: slru.h:36
int num_files
Definition: slru.c:77
LWLock lock
Definition: lwlock.h:79
static bool SlruPhysicalReadPage(SlruCtl ctl, int pageno, int slotno)
Definition: slru.c:639
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
int * page_lru_count
Definition: slru.h:70
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:670
#define StrNCpy(dst, src, len)
Definition: c.h:826
bool SlruScanDirectory(SlruCtl ctl, SlruScanCallback callback, void *data)
Definition: slru.c:1376
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2439
size_t Size
Definition: c.h:404
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: pgstat.h:1220
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1117
#define MAXALIGN(LEN)
Definition: c.h:623
int num_slots
Definition: slru.h:60
const char * name
Definition: encode.c:521
struct SlruFlushData SlruFlushData
static int SlruSelectLRUPage(SlruCtl ctl, int pageno)
Definition: slru.c:966
static int slru_errno
Definition: slru.c:124
static char * filename
Definition: pg_dumpall.c:90
int errmsg(const char *fmt,...)
Definition: elog.c:797
bool * page_dirty
Definition: slru.h:68
int i
SlruShared shared
Definition: slru.h:115
#define BUFFERALIGN(LEN)
Definition: c.h:625
int lwlock_tranche_id
Definition: slru.h:102
bool(* PagePrecedes)(int, int)
Definition: slru.h:128
void SlruDeleteSegment(SlruCtl ctl, int segno)
Definition: slru.c:1261
int pg_fsync(int fd)
Definition: fd.c:338
char d_name[MAX_PATH]
Definition: dirent.h:14
#define elog
Definition: elog.h:219
#define SLRU_PAGES_PER_SEGMENT
Definition: slru.h:33
int SimpleLruZeroPage(SlruCtl ctl, int pageno)
Definition: slru.c:263
#define read(a, b, c)
Definition: win32.h:13
int FreeDir(DIR *dir)
Definition: fd.c:2482
void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns, LWLock *ctllock, const char *subdir, int tranche_id)
Definition: slru.c:165
#define SlruRecentlyUsed(shared, slotno)
Definition: slru.c:103
int fd[MAX_FLUSH_BUFFERS]
Definition: slru.c:78