PostgreSQL Source Code  git master
buffile.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * buffile.c
4  * Management of large buffered temporary files.
5  *
6  * Portions Copyright (c) 1996-2021, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  * IDENTIFICATION
10  * src/backend/storage/file/buffile.c
11  *
12  * NOTES:
13  *
14  * BufFiles provide a very incomplete emulation of stdio atop virtual Files
15  * (as managed by fd.c). Currently, we only support the buffered-I/O
16  * aspect of stdio: a read or write of the low-level File occurs only
17  * when the buffer is filled or emptied. This is an even bigger win
18  * for virtual Files than for ordinary kernel files, since reducing the
19  * frequency with which a virtual File is touched reduces "thrashing"
20  * of opening/closing file descriptors.
21  *
22  * Note that BufFile structs are allocated with palloc(), and therefore
23  * will go away automatically at query/transaction end. Since the underlying
24  * virtual Files are made with OpenTemporaryFile, all resources for
25  * the file are certain to be cleaned up even if processing is aborted
26  * by ereport(ERROR). The data structures required are made in the
27  * palloc context that was current when the BufFile was created, and
28  * any external resources such as temp files are owned by the ResourceOwner
29  * that was current at that time.
30  *
31  * BufFile also supports temporary files that exceed the OS file size limit
32  * (by opening multiple fd.c temporary files). This is an essential feature
33  * for sorts and hashjoins on large amounts of data.
34  *
35  * BufFile supports temporary files that can be shared with other backends, as
36  * infrastructure for parallel execution. Such files need to be created as a
37  * member of a SharedFileSet that all participants are attached to.
38  *
39  * BufFile also supports temporary files that can be used by the single backend
40  * when the corresponding files need to be survived across the transaction and
41  * need to be opened and closed multiple times. Such files need to be created
42  * as a member of a FileSet.
43  *-------------------------------------------------------------------------
44  */
45 
46 #include "postgres.h"
47 
48 #include "commands/tablespace.h"
49 #include "executor/instrument.h"
50 #include "miscadmin.h"
51 #include "pgstat.h"
52 #include "storage/buf_internals.h"
53 #include "storage/buffile.h"
54 #include "storage/fd.h"
55 #include "utils/resowner.h"
56 
57 /*
58  * We break BufFiles into gigabyte-sized segments, regardless of RELSEG_SIZE.
59  * The reason is that we'd like large BufFiles to be spread across multiple
60  * tablespaces when available.
61  */
62 #define MAX_PHYSICAL_FILESIZE 0x40000000
63 #define BUFFILE_SEG_SIZE (MAX_PHYSICAL_FILESIZE / BLCKSZ)
64 
65 /*
66  * This data structure represents a buffered file that consists of one or
67  * more physical files (each accessed through a virtual file descriptor
68  * managed by fd.c).
69  */
70 struct BufFile
71 {
72  int numFiles; /* number of physical files in set */
73  /* all files except the last have length exactly MAX_PHYSICAL_FILESIZE */
74  File *files; /* palloc'd array with numFiles entries */
75 
76  bool isInterXact; /* keep open over transactions? */
77  bool dirty; /* does buffer need to be written? */
78  bool readOnly; /* has the file been set to read only? */
79 
80  FileSet *fileset; /* space for fileset based segment files */
81  const char *name; /* name of fileset based BufFile */
82 
83  /*
84  * resowner is the ResourceOwner to use for underlying temp files. (We
85  * don't need to remember the memory context we're using explicitly,
86  * because after creation we only repalloc our arrays larger.)
87  */
89 
90  /*
91  * "current pos" is position of start of buffer within the logical file.
92  * Position as seen by user of BufFile is (curFile, curOffset + pos).
93  */
94  int curFile; /* file index (0..n) part of current pos */
95  off_t curOffset; /* offset part of current pos */
96  int pos; /* next read/write position in buffer */
97  int nbytes; /* total # of valid bytes in buffer */
99 };
100 
101 static BufFile *makeBufFileCommon(int nfiles);
102 static BufFile *makeBufFile(File firstfile);
103 static void extendBufFile(BufFile *file);
104 static void BufFileLoadBuffer(BufFile *file);
105 static void BufFileDumpBuffer(BufFile *file);
106 static void BufFileFlush(BufFile *file);
107 static File MakeNewFileSetSegment(BufFile *file, int segment);
108 
109 /*
110  * Create BufFile and perform the common initialization.
111  */
112 static BufFile *
113 makeBufFileCommon(int nfiles)
114 {
115  BufFile *file = (BufFile *) palloc(sizeof(BufFile));
116 
117  file->numFiles = nfiles;
118  file->isInterXact = false;
119  file->dirty = false;
121  file->curFile = 0;
122  file->curOffset = 0L;
123  file->pos = 0;
124  file->nbytes = 0;
125 
126  return file;
127 }
128 
129 /*
130  * Create a BufFile given the first underlying physical file.
131  * NOTE: caller must set isInterXact if appropriate.
132  */
133 static BufFile *
134 makeBufFile(File firstfile)
135 {
136  BufFile *file = makeBufFileCommon(1);
137 
138  file->files = (File *) palloc(sizeof(File));
139  file->files[0] = firstfile;
140  file->readOnly = false;
141  file->fileset = NULL;
142  file->name = NULL;
143 
144  return file;
145 }
146 
147 /*
148  * Add another component temp file.
149  */
150 static void
152 {
153  File pfile;
154  ResourceOwner oldowner;
155 
156  /* Be sure to associate the file with the BufFile's resource owner */
157  oldowner = CurrentResourceOwner;
159 
160  if (file->fileset == NULL)
161  pfile = OpenTemporaryFile(file->isInterXact);
162  else
163  pfile = MakeNewFileSetSegment(file, file->numFiles);
164 
165  Assert(pfile >= 0);
166 
167  CurrentResourceOwner = oldowner;
168 
169  file->files = (File *) repalloc(file->files,
170  (file->numFiles + 1) * sizeof(File));
171  file->files[file->numFiles] = pfile;
172  file->numFiles++;
173 }
174 
175 /*
176  * Create a BufFile for a new temporary file (which will expand to become
177  * multiple temporary files if more than MAX_PHYSICAL_FILESIZE bytes are
178  * written to it).
179  *
180  * If interXact is true, the temp file will not be automatically deleted
181  * at end of transaction.
182  *
183  * Note: if interXact is true, the caller had better be calling us in a
184  * memory context, and with a resource owner, that will survive across
185  * transaction boundaries.
186  */
187 BufFile *
188 BufFileCreateTemp(bool interXact)
189 {
190  BufFile *file;
191  File pfile;
192 
193  /*
194  * Ensure that temp tablespaces are set up for OpenTemporaryFile to use.
195  * Possibly the caller will have done this already, but it seems useful to
196  * double-check here. Failure to do this at all would result in the temp
197  * files always getting placed in the default tablespace, which is a
198  * pretty hard-to-detect bug. Callers may prefer to do it earlier if they
199  * want to be sure that any required catalog access is done in some other
200  * resource context.
201  */
203 
204  pfile = OpenTemporaryFile(interXact);
205  Assert(pfile >= 0);
206 
207  file = makeBufFile(pfile);
208  file->isInterXact = interXact;
209 
210  return file;
211 }
212 
213 /*
214  * Build the name for a given segment of a given BufFile.
215  */
216 static void
217 FileSetSegmentName(char *name, const char *buffile_name, int segment)
218 {
219  snprintf(name, MAXPGPATH, "%s.%d", buffile_name, segment);
220 }
221 
222 /*
223  * Create a new segment file backing a fileset based BufFile.
224  */
225 static File
226 MakeNewFileSetSegment(BufFile *buffile, int segment)
227 {
228  char name[MAXPGPATH];
229  File file;
230 
231  /*
232  * It is possible that there are files left over from before a crash
233  * restart with the same name. In order for BufFileOpenFileSet() not to
234  * get confused about how many segments there are, we'll unlink the next
235  * segment number if it already exists.
236  */
237  FileSetSegmentName(name, buffile->name, segment + 1);
238  FileSetDelete(buffile->fileset, name, true);
239 
240  /* Create the new segment. */
241  FileSetSegmentName(name, buffile->name, segment);
242  file = FileSetCreate(buffile->fileset, name);
243 
244  /* FileSetCreate would've errored out */
245  Assert(file > 0);
246 
247  return file;
248 }
249 
250 /*
251  * Create a BufFile that can be discovered and opened read-only by other
252  * backends that are attached to the same SharedFileSet using the same name.
253  *
254  * The naming scheme for fileset based BufFiles is left up to the calling code.
255  * The name will appear as part of one or more filenames on disk, and might
256  * provide clues to administrators about which subsystem is generating
257  * temporary file data. Since each SharedFileSet object is backed by one or
258  * more uniquely named temporary directory, names don't conflict with
259  * unrelated SharedFileSet objects.
260  */
261 BufFile *
263 {
264  BufFile *file;
265 
266  file = makeBufFileCommon(1);
267  file->fileset = fileset;
268  file->name = pstrdup(name);
269  file->files = (File *) palloc(sizeof(File));
270  file->files[0] = MakeNewFileSetSegment(file, 0);
271  file->readOnly = false;
272 
273  return file;
274 }
275 
276 /*
277  * Open a file that was previously created in another backend (or this one)
278  * with BufFileCreateFileSet in the same FileSet using the same name.
279  * The backend that created the file must have called BufFileClose() or
280  * BufFileExportFileSet() to make sure that it is ready to be opened by other
281  * backends and render it read-only. If missing_ok is true, which indicates
282  * that missing files can be safely ignored, then return NULL if the BufFile
283  * with the given name is not found, otherwise, throw an error.
284  */
285 BufFile *
287  bool missing_ok)
288 {
289  BufFile *file;
290  char segment_name[MAXPGPATH];
291  Size capacity = 16;
292  File *files;
293  int nfiles = 0;
294 
295  files = palloc(sizeof(File) * capacity);
296 
297  /*
298  * We don't know how many segments there are, so we'll probe the
299  * filesystem to find out.
300  */
301  for (;;)
302  {
303  /* See if we need to expand our file segment array. */
304  if (nfiles + 1 > capacity)
305  {
306  capacity *= 2;
307  files = repalloc(files, sizeof(File) * capacity);
308  }
309  /* Try to load a segment. */
310  FileSetSegmentName(segment_name, name, nfiles);
311  files[nfiles] = FileSetOpen(fileset, segment_name, mode);
312  if (files[nfiles] <= 0)
313  break;
314  ++nfiles;
315 
317  }
318 
319  /*
320  * If we didn't find any files at all, then no BufFile exists with this
321  * name.
322  */
323  if (nfiles == 0)
324  {
325  /* free the memory */
326  pfree(files);
327 
328  if (missing_ok)
329  return NULL;
330 
331  ereport(ERROR,
333  errmsg("could not open temporary file \"%s\" from BufFile \"%s\": %m",
334  segment_name, name)));
335  }
336 
337  file = makeBufFileCommon(nfiles);
338  file->files = files;
339  file->readOnly = (mode == O_RDONLY);
340  file->fileset = fileset;
341  file->name = pstrdup(name);
342 
343  return file;
344 }
345 
346 /*
347  * Delete a BufFile that was created by BufFileCreateFileSet in the given
348  * FileSet using the given name.
349  *
350  * It is not necessary to delete files explicitly with this function. It is
351  * provided only as a way to delete files proactively, rather than waiting for
352  * the FileSet to be cleaned up.
353  *
354  * Only one backend should attempt to delete a given name, and should know
355  * that it exists and has been exported or closed otherwise missing_ok should
356  * be passed true.
357  */
358 void
359 BufFileDeleteFileSet(FileSet *fileset, const char *name, bool missing_ok)
360 {
361  char segment_name[MAXPGPATH];
362  int segment = 0;
363  bool found = false;
364 
365  /*
366  * We don't know how many segments the file has. We'll keep deleting
367  * until we run out. If we don't manage to find even an initial segment,
368  * raise an error.
369  */
370  for (;;)
371  {
372  FileSetSegmentName(segment_name, name, segment);
373  if (!FileSetDelete(fileset, segment_name, true))
374  break;
375  found = true;
376  ++segment;
377 
379  }
380 
381  if (!found && !missing_ok)
382  elog(ERROR, "could not delete unknown BufFile \"%s\"", name);
383 }
384 
385 /*
386  * BufFileExportFileSet --- flush and make read-only, in preparation for sharing.
387  */
388 void
390 {
391  /* Must be a file belonging to a FileSet. */
392  Assert(file->fileset != NULL);
393 
394  /* It's probably a bug if someone calls this twice. */
395  Assert(!file->readOnly);
396 
397  BufFileFlush(file);
398  file->readOnly = true;
399 }
400 
401 /*
402  * Close a BufFile
403  *
404  * Like fclose(), this also implicitly FileCloses the underlying File.
405  */
406 void
408 {
409  int i;
410 
411  /* flush any unwritten data */
412  BufFileFlush(file);
413  /* close and delete the underlying file(s) */
414  for (i = 0; i < file->numFiles; i++)
415  FileClose(file->files[i]);
416  /* release the buffer space */
417  pfree(file->files);
418  pfree(file);
419 }
420 
421 /*
422  * BufFileLoadBuffer
423  *
424  * Load some data into buffer, if possible, starting from curOffset.
425  * At call, must have dirty = false, pos and nbytes = 0.
426  * On exit, nbytes is number of bytes loaded.
427  */
428 static void
430 {
431  File thisfile;
432 
433  /*
434  * Advance to next component file if necessary and possible.
435  */
436  if (file->curOffset >= MAX_PHYSICAL_FILESIZE &&
437  file->curFile + 1 < file->numFiles)
438  {
439  file->curFile++;
440  file->curOffset = 0L;
441  }
442 
443  /*
444  * Read whatever we can get, up to a full bufferload.
445  */
446  thisfile = file->files[file->curFile];
447  file->nbytes = FileRead(thisfile,
448  file->buffer.data,
449  sizeof(file->buffer),
450  file->curOffset,
452  if (file->nbytes < 0)
453  {
454  file->nbytes = 0;
455  ereport(ERROR,
457  errmsg("could not read file \"%s\": %m",
458  FilePathName(thisfile))));
459  }
460 
461  /* we choose not to advance curOffset here */
462 
463  if (file->nbytes > 0)
465 }
466 
467 /*
468  * BufFileDumpBuffer
469  *
470  * Dump buffer contents starting at curOffset.
471  * At call, should have dirty = true, nbytes > 0.
472  * On exit, dirty is cleared if successful write, and curOffset is advanced.
473  */
474 static void
476 {
477  int wpos = 0;
478  int bytestowrite;
479  File thisfile;
480 
481  /*
482  * Unlike BufFileLoadBuffer, we must dump the whole buffer even if it
483  * crosses a component-file boundary; so we need a loop.
484  */
485  while (wpos < file->nbytes)
486  {
487  off_t availbytes;
488 
489  /*
490  * Advance to next component file if necessary and possible.
491  */
492  if (file->curOffset >= MAX_PHYSICAL_FILESIZE)
493  {
494  while (file->curFile + 1 >= file->numFiles)
495  extendBufFile(file);
496  file->curFile++;
497  file->curOffset = 0L;
498  }
499 
500  /*
501  * Determine how much we need to write into this file.
502  */
503  bytestowrite = file->nbytes - wpos;
504  availbytes = MAX_PHYSICAL_FILESIZE - file->curOffset;
505 
506  if ((off_t) bytestowrite > availbytes)
507  bytestowrite = (int) availbytes;
508 
509  thisfile = file->files[file->curFile];
510  bytestowrite = FileWrite(thisfile,
511  file->buffer.data + wpos,
512  bytestowrite,
513  file->curOffset,
515  if (bytestowrite <= 0)
516  ereport(ERROR,
518  errmsg("could not write to file \"%s\": %m",
519  FilePathName(thisfile))));
520  file->curOffset += bytestowrite;
521  wpos += bytestowrite;
522 
524  }
525  file->dirty = false;
526 
527  /*
528  * At this point, curOffset has been advanced to the end of the buffer,
529  * ie, its original value + nbytes. We need to make it point to the
530  * logical file position, ie, original value + pos, in case that is less
531  * (as could happen due to a small backwards seek in a dirty buffer!)
532  */
533  file->curOffset -= (file->nbytes - file->pos);
534  if (file->curOffset < 0) /* handle possible segment crossing */
535  {
536  file->curFile--;
537  Assert(file->curFile >= 0);
539  }
540 
541  /*
542  * Now we can set the buffer empty without changing the logical position
543  */
544  file->pos = 0;
545  file->nbytes = 0;
546 }
547 
548 /*
549  * BufFileRead
550  *
551  * Like fread() except we assume 1-byte element size and report I/O errors via
552  * ereport().
553  */
554 size_t
555 BufFileRead(BufFile *file, void *ptr, size_t size)
556 {
557  size_t nread = 0;
558  size_t nthistime;
559 
560  BufFileFlush(file);
561 
562  while (size > 0)
563  {
564  if (file->pos >= file->nbytes)
565  {
566  /* Try to load more data into buffer. */
567  file->curOffset += file->pos;
568  file->pos = 0;
569  file->nbytes = 0;
570  BufFileLoadBuffer(file);
571  if (file->nbytes <= 0)
572  break; /* no more data available */
573  }
574 
575  nthistime = file->nbytes - file->pos;
576  if (nthistime > size)
577  nthistime = size;
578  Assert(nthistime > 0);
579 
580  memcpy(ptr, file->buffer.data + file->pos, nthistime);
581 
582  file->pos += nthistime;
583  ptr = (void *) ((char *) ptr + nthistime);
584  size -= nthistime;
585  nread += nthistime;
586  }
587 
588  return nread;
589 }
590 
591 /*
592  * BufFileWrite
593  *
594  * Like fwrite() except we assume 1-byte element size and report errors via
595  * ereport().
596  */
597 void
598 BufFileWrite(BufFile *file, void *ptr, size_t size)
599 {
600  size_t nthistime;
601 
602  Assert(!file->readOnly);
603 
604  while (size > 0)
605  {
606  if (file->pos >= BLCKSZ)
607  {
608  /* Buffer full, dump it out */
609  if (file->dirty)
610  BufFileDumpBuffer(file);
611  else
612  {
613  /* Hmm, went directly from reading to writing? */
614  file->curOffset += file->pos;
615  file->pos = 0;
616  file->nbytes = 0;
617  }
618  }
619 
620  nthistime = BLCKSZ - file->pos;
621  if (nthistime > size)
622  nthistime = size;
623  Assert(nthistime > 0);
624 
625  memcpy(file->buffer.data + file->pos, ptr, nthistime);
626 
627  file->dirty = true;
628  file->pos += nthistime;
629  if (file->nbytes < file->pos)
630  file->nbytes = file->pos;
631  ptr = (void *) ((char *) ptr + nthistime);
632  size -= nthistime;
633  }
634 }
635 
636 /*
637  * BufFileFlush
638  *
639  * Like fflush(), except that I/O errors are reported with ereport().
640  */
641 static void
643 {
644  if (file->dirty)
645  BufFileDumpBuffer(file);
646 
647  Assert(!file->dirty);
648 }
649 
650 /*
651  * BufFileSeek
652  *
653  * Like fseek(), except that target position needs two values in order to
654  * work when logical filesize exceeds maximum value representable by off_t.
655  * We do not support relative seeks across more than that, however.
656  * I/O errors are reported by ereport().
657  *
658  * Result is 0 if OK, EOF if not. Logical position is not moved if an
659  * impossible seek is attempted.
660  */
661 int
662 BufFileSeek(BufFile *file, int fileno, off_t offset, int whence)
663 {
664  int newFile;
665  off_t newOffset;
666 
667  switch (whence)
668  {
669  case SEEK_SET:
670  if (fileno < 0)
671  return EOF;
672  newFile = fileno;
673  newOffset = offset;
674  break;
675  case SEEK_CUR:
676 
677  /*
678  * Relative seek considers only the signed offset, ignoring
679  * fileno. Note that large offsets (> 1 GB) risk overflow in this
680  * add, unless we have 64-bit off_t.
681  */
682  newFile = file->curFile;
683  newOffset = (file->curOffset + file->pos) + offset;
684  break;
685  case SEEK_END:
686 
687  /*
688  * The file size of the last file gives us the end offset of that
689  * file.
690  */
691  newFile = file->numFiles - 1;
692  newOffset = FileSize(file->files[file->numFiles - 1]);
693  if (newOffset < 0)
694  ereport(ERROR,
696  errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
697  FilePathName(file->files[file->numFiles - 1]),
698  file->name)));
699  break;
700  default:
701  elog(ERROR, "invalid whence: %d", whence);
702  return EOF;
703  }
704  while (newOffset < 0)
705  {
706  if (--newFile < 0)
707  return EOF;
708  newOffset += MAX_PHYSICAL_FILESIZE;
709  }
710  if (newFile == file->curFile &&
711  newOffset >= file->curOffset &&
712  newOffset <= file->curOffset + file->nbytes)
713  {
714  /*
715  * Seek is to a point within existing buffer; we can just adjust
716  * pos-within-buffer, without flushing buffer. Note this is OK
717  * whether reading or writing, but buffer remains dirty if we were
718  * writing.
719  */
720  file->pos = (int) (newOffset - file->curOffset);
721  return 0;
722  }
723  /* Otherwise, must reposition buffer, so flush any dirty data */
724  BufFileFlush(file);
725 
726  /*
727  * At this point and no sooner, check for seek past last segment. The
728  * above flush could have created a new segment, so checking sooner would
729  * not work (at least not with this code).
730  */
731 
732  /* convert seek to "start of next seg" to "end of last seg" */
733  if (newFile == file->numFiles && newOffset == 0)
734  {
735  newFile--;
736  newOffset = MAX_PHYSICAL_FILESIZE;
737  }
738  while (newOffset > MAX_PHYSICAL_FILESIZE)
739  {
740  if (++newFile >= file->numFiles)
741  return EOF;
742  newOffset -= MAX_PHYSICAL_FILESIZE;
743  }
744  if (newFile >= file->numFiles)
745  return EOF;
746  /* Seek is OK! */
747  file->curFile = newFile;
748  file->curOffset = newOffset;
749  file->pos = 0;
750  file->nbytes = 0;
751  return 0;
752 }
753 
754 void
755 BufFileTell(BufFile *file, int *fileno, off_t *offset)
756 {
757  *fileno = file->curFile;
758  *offset = file->curOffset + file->pos;
759 }
760 
761 /*
762  * BufFileSeekBlock --- block-oriented seek
763  *
764  * Performs absolute seek to the start of the n'th BLCKSZ-sized block of
765  * the file. Note that users of this interface will fail if their files
766  * exceed BLCKSZ * LONG_MAX bytes, but that is quite a lot; we don't work
767  * with tables bigger than that, either...
768  *
769  * Result is 0 if OK, EOF if not. Logical position is not moved if an
770  * impossible seek is attempted.
771  */
772 int
773 BufFileSeekBlock(BufFile *file, long blknum)
774 {
775  return BufFileSeek(file,
776  (int) (blknum / BUFFILE_SEG_SIZE),
777  (off_t) (blknum % BUFFILE_SEG_SIZE) * BLCKSZ,
778  SEEK_SET);
779 }
780 
781 #ifdef NOT_USED
782 /*
783  * BufFileTellBlock --- block-oriented tell
784  *
785  * Any fractional part of a block in the current seek position is ignored.
786  */
787 long
788 BufFileTellBlock(BufFile *file)
789 {
790  long blknum;
791 
792  blknum = (file->curOffset + file->pos) / BLCKSZ;
793  blknum += file->curFile * BUFFILE_SEG_SIZE;
794  return blknum;
795 }
796 
797 #endif
798 
799 /*
800  * Return the current fileset based BufFile size.
801  *
802  * Counts any holes left behind by BufFileAppend as part of the size.
803  * ereport()s on failure.
804  */
805 int64
807 {
808  int64 lastFileSize;
809 
810  Assert(file->fileset != NULL);
811 
812  /* Get the size of the last physical file. */
813  lastFileSize = FileSize(file->files[file->numFiles - 1]);
814  if (lastFileSize < 0)
815  ereport(ERROR,
817  errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
818  FilePathName(file->files[file->numFiles - 1]),
819  file->name)));
820 
821  return ((file->numFiles - 1) * (int64) MAX_PHYSICAL_FILESIZE) +
822  lastFileSize;
823 }
824 
825 /*
826  * Append the contents of source file (managed within fileset) to
827  * end of target file (managed within same fileset).
828  *
829  * Note that operation subsumes ownership of underlying resources from
830  * "source". Caller should never call BufFileClose against source having
831  * called here first. Resource owners for source and target must match,
832  * too.
833  *
834  * This operation works by manipulating lists of segment files, so the
835  * file content is always appended at a MAX_PHYSICAL_FILESIZE-aligned
836  * boundary, typically creating empty holes before the boundary. These
837  * areas do not contain any interesting data, and cannot be read from by
838  * caller.
839  *
840  * Returns the block number within target where the contents of source
841  * begins. Caller should apply this as an offset when working off block
842  * positions that are in terms of the original BufFile space.
843  */
844 long
846 {
847  long startBlock = target->numFiles * BUFFILE_SEG_SIZE;
848  int newNumFiles = target->numFiles + source->numFiles;
849  int i;
850 
851  Assert(target->fileset != NULL);
852  Assert(source->readOnly);
853  Assert(!source->dirty);
854  Assert(source->fileset != NULL);
855 
856  if (target->resowner != source->resowner)
857  elog(ERROR, "could not append BufFile with non-matching resource owner");
858 
859  target->files = (File *)
860  repalloc(target->files, sizeof(File) * newNumFiles);
861  for (i = target->numFiles; i < newNumFiles; i++)
862  target->files[i] = source->files[i - target->numFiles];
863  target->numFiles = newNumFiles;
864 
865  return startBlock;
866 }
867 
868 /*
869  * Truncate a BufFile created by BufFileCreateFileSet up to the given fileno
870  * and the offset.
871  */
872 void
873 BufFileTruncateFileSet(BufFile *file, int fileno, off_t offset)
874 {
875  int numFiles = file->numFiles;
876  int newFile = fileno;
877  off_t newOffset = file->curOffset;
878  char segment_name[MAXPGPATH];
879  int i;
880 
881  /*
882  * Loop over all the files up to the given fileno and remove the files
883  * that are greater than the fileno and truncate the given file up to the
884  * offset. Note that we also remove the given fileno if the offset is 0
885  * provided it is not the first file in which we truncate it.
886  */
887  for (i = file->numFiles - 1; i >= fileno; i--)
888  {
889  if ((i != fileno || offset == 0) && i != 0)
890  {
891  FileSetSegmentName(segment_name, file->name, i);
892  FileClose(file->files[i]);
893  if (!FileSetDelete(file->fileset, segment_name, true))
894  ereport(ERROR,
896  errmsg("could not delete fileset \"%s\": %m",
897  segment_name)));
898  numFiles--;
899  newOffset = MAX_PHYSICAL_FILESIZE;
900 
901  /*
902  * This is required to indicate that we have deleted the given
903  * fileno.
904  */
905  if (i == fileno)
906  newFile--;
907  }
908  else
909  {
910  if (FileTruncate(file->files[i], offset,
912  ereport(ERROR,
914  errmsg("could not truncate file \"%s\": %m",
915  FilePathName(file->files[i]))));
916  newOffset = offset;
917  }
918  }
919 
920  file->numFiles = numFiles;
921 
922  /*
923  * If the truncate point is within existing buffer then we can just adjust
924  * pos within buffer.
925  */
926  if (newFile == file->curFile &&
927  newOffset >= file->curOffset &&
928  newOffset <= file->curOffset + file->nbytes)
929  {
930  /* No need to reset the current pos if the new pos is greater. */
931  if (newOffset <= file->curOffset + file->pos)
932  file->pos = (int) (newOffset - file->curOffset);
933 
934  /* Adjust the nbytes for the current buffer. */
935  file->nbytes = (int) (newOffset - file->curOffset);
936  }
937  else if (newFile == file->curFile &&
938  newOffset < file->curOffset)
939  {
940  /*
941  * The truncate point is within the existing file but prior to the
942  * current position, so we can forget the current buffer and reset the
943  * current position.
944  */
945  file->curOffset = newOffset;
946  file->pos = 0;
947  file->nbytes = 0;
948  }
949  else if (newFile < file->curFile)
950  {
951  /*
952  * The truncate point is prior to the current file, so need to reset
953  * the current position accordingly.
954  */
955  file->curFile = newFile;
956  file->curOffset = newOffset;
957  file->pos = 0;
958  file->nbytes = 0;
959  }
960  /* Nothing to do, if the truncate point is beyond current file. */
961 }
static PgChecksumMode mode
Definition: pg_checksums.c:65
static void BufFileFlush(BufFile *file)
Definition: buffile.c:642
int nbytes
Definition: buffile.c:97
int BufFileSeek(BufFile *file, int fileno, off_t offset, int whence)
Definition: buffile.c:662
File FileSetOpen(FileSet *fileset, const char *name, int mode)
Definition: fileset.c:121
ResourceOwner CurrentResourceOwner
Definition: resowner.c:146
char * pstrdup(const char *in)
Definition: mcxt.c:1299
int64 BufFileSize(BufFile *file)
Definition: buffile.c:806
void BufFileTruncateFileSet(BufFile *file, int fileno, off_t offset)
Definition: buffile.c:873
off_t FileSize(File file)
Definition: fd.c:2285
BufFile * BufFileOpenFileSet(FileSet *fileset, const char *name, int mode, bool missing_ok)
Definition: buffile.c:286
#define MAX_PHYSICAL_FILESIZE
Definition: buffile.c:62
bool readOnly
Definition: buffile.c:78
PGAlignedBlock buffer
Definition: buffile.c:98
#define BUFFILE_SEG_SIZE
Definition: buffile.c:63
int64 temp_blks_written
Definition: instrument.h:35
static void extendBufFile(BufFile *file)
Definition: buffile.c:151
int numFiles
Definition: buffile.c:72
char * FilePathName(File file)
Definition: fd.c:2337
ResourceOwner resowner
Definition: buffile.c:88
static void FileSetSegmentName(char *name, const char *buffile_name, int segment)
Definition: buffile.c:217
FileSet * fileset
Definition: buffile.c:80
void BufFileDeleteFileSet(FileSet *fileset, const char *name, bool missing_ok)
Definition: buffile.c:359
void BufFileClose(BufFile *file)
Definition: buffile.c:407
File FileSetCreate(FileSet *fileset, const char *name)
Definition: fileset.c:94
char data[BLCKSZ]
Definition: c.h:1141
void pfree(void *pointer)
Definition: mcxt.c:1169
void BufFileExportFileSet(BufFile *file)
Definition: buffile.c:389
#define ERROR
Definition: elog.h:46
void PrepareTempTablespaces(void)
Definition: tablespace.c:1332
BufFile * BufFileCreateTemp(bool interXact)
Definition: buffile.c:188
#define MAXPGPATH
bool isInterXact
Definition: buffile.c:76
int errcode_for_file_access(void)
Definition: elog.c:721
static BufFile * makeBufFile(File firstfile)
Definition: buffile.c:134
static void BufFileDumpBuffer(BufFile *file)
Definition: buffile.c:475
File * files
Definition: buffile.c:74
#define wpos(wep)
Definition: tsrank.c:26
off_t curOffset
Definition: buffile.c:95
const char * name
Definition: buffile.c:81
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2166
int BufFileSeekBlock(BufFile *file, long blknum)
Definition: buffile.c:773
int curFile
Definition: buffile.c:94
#define ereport(elevel,...)
Definition: elog.h:157
void BufFileTell(BufFile *file, int *fileno, off_t *offset)
Definition: buffile.c:755
void FileClose(File file)
Definition: fd.c:1959
static File MakeNewFileSetSegment(BufFile *file, int segment)
Definition: buffile.c:226
BufFile * BufFileCreateFileSet(FileSet *fileset, const char *name)
Definition: buffile.c:262
#define Assert(condition)
Definition: c.h:804
bool FileSetDelete(FileSet *fileset, const char *name, bool error_on_failure)
Definition: fileset.c:138
File OpenTemporaryFile(bool interXact)
Definition: fd.c:1705
static rewind_source * source
Definition: pg_rewind.c:79
size_t Size
Definition: c.h:540
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1182
bool dirty
Definition: buffile.c:77
void * palloc(Size size)
Definition: mcxt.c:1062
int errmsg(const char *fmt,...)
Definition: elog.c:909
#define elog(elevel,...)
Definition: elog.h:232
int i
size_t BufFileRead(BufFile *file, void *ptr, size_t size)
Definition: buffile.c:555
static void BufFileLoadBuffer(BufFile *file)
Definition: buffile.c:429
int64 temp_blks_read
Definition: instrument.h:34
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:120
void BufFileWrite(BufFile *file, void *ptr, size_t size)
Definition: buffile.c:598
long BufFileAppend(BufFile *target, BufFile *source)
Definition: buffile.c:845
int FileRead(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2110
#define snprintf
Definition: port.h:216
int FileTruncate(File file, off_t offset, uint32 wait_event_info)
Definition: fd.c:2302
int File
Definition: fd.h:54
BufferUsage pgBufferUsage
Definition: instrument.c:20
int pos
Definition: buffile.c:96
static BufFile * makeBufFileCommon(int nfiles)
Definition: buffile.c:113