1 /*-------------------------------------------------------------------------
2 *
3 * buffile.c
4 * Management of large buffered temporary files.
5 *
6 * Portions Copyright (c) 1996-2018, 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 made read-only and shared with
36 * other backends, as infrastructure for parallel execution. Such files need
37 * to be created as a member of a SharedFileSet that all participants are
38 * attached to.
39 *-------------------------------------------------------------------------
40 */
41
42 #include "postgres.h"
43
44 #include "executor/instrument.h"
45 #include "miscadmin.h"
46 #include "pgstat.h"
47 #include "storage/fd.h"
48 #include "storage/buffile.h"
49 #include "storage/buf_internals.h"
50 #include "utils/resowner.h"
51
52 /*
53 * We break BufFiles into gigabyte-sized segments, regardless of RELSEG_SIZE.
54 * The reason is that we'd like large BufFiles to be spread across multiple
55 * tablespaces when available.
56 */
57 #define MAX_PHYSICAL_FILESIZE 0x40000000
58 #define BUFFILE_SEG_SIZE (MAX_PHYSICAL_FILESIZE / BLCKSZ)
59
60 /*
61 * This data structure represents a buffered file that consists of one or
62 * more physical files (each accessed through a virtual file descriptor
63 * managed by fd.c).
64 */
65 struct BufFile
66 {
67 int numFiles; /* number of physical files in set */
68 /* all files except the last have length exactly MAX_PHYSICAL_FILESIZE */
69 File *files; /* palloc'd array with numFiles entries */
70 off_t *offsets; /* palloc'd array with numFiles entries */
71
72 /*
73 * offsets[i] is the current seek position of files[i]. We use this to
74 * avoid making redundant FileSeek calls.
75 */
76
77 bool isInterXact; /* keep open over transactions? */
78 bool dirty; /* does buffer need to be written? */
79 bool readOnly; /* has the file been set to read only? */
80
81 SharedFileSet *fileset; /* space for segment files if shared */
82 const char *name; /* name of this BufFile if shared */
83
84 /*
85 * resowner is the ResourceOwner to use for underlying temp files. (We
86 * don't need to remember the memory context we're using explicitly,
87 * because after creation we only repalloc our arrays larger.)
88 */
89 ResourceOwner resowner;
90
91 /*
92 * "current pos" is position of start of buffer within the logical file.
93 * Position as seen by user of BufFile is (curFile, curOffset + pos).
94 */
95 int curFile; /* file index (0..n) part of current pos */
96 off_t curOffset; /* offset part of current pos */
97 int pos; /* next read/write position in buffer */
98 int nbytes; /* total # of valid bytes in buffer */
99 PGAlignedBlock buffer;
100 };
101
102 static BufFile *makeBufFileCommon(int nfiles);
103 static BufFile *makeBufFile(File firstfile);
104 static void extendBufFile(BufFile *file);
105 static void BufFileLoadBuffer(BufFile *file);
106 static void BufFileDumpBuffer(BufFile *file);
107 static void BufFileFlush(BufFile *file);
108 static File MakeNewSharedSegment(BufFile *file, int segment);
109
110 /*
111 * Create BufFile and perform the common initialization.
112 */
113 static BufFile *
makeBufFileCommon(int nfiles)114 makeBufFileCommon(int nfiles)
115 {
116 BufFile *file = (BufFile *) palloc(sizeof(BufFile));
117
118 file->numFiles = nfiles;
119 file->offsets = (off_t *) palloc0(sizeof(off_t) * nfiles);
120 file->isInterXact = false;
121 file->dirty = false;
122 file->resowner = CurrentResourceOwner;
123 file->curFile = 0;
124 file->curOffset = 0L;
125 file->pos = 0;
126 file->nbytes = 0;
127
128 return file;
129 }
130
131 /*
132 * Create a BufFile given the first underlying physical file.
133 * NOTE: caller must set isInterXact if appropriate.
134 */
135 static BufFile *
makeBufFile(File firstfile)136 makeBufFile(File firstfile)
137 {
138 BufFile *file = makeBufFileCommon(1);
139
140 file->files = (File *) palloc(sizeof(File));
141 file->files[0] = firstfile;
142 file->readOnly = false;
143 file->fileset = NULL;
144 file->name = NULL;
145
146 return file;
147 }
148
149 /*
150 * Add another component temp file.
151 */
152 static void
extendBufFile(BufFile * file)153 extendBufFile(BufFile *file)
154 {
155 File pfile;
156 ResourceOwner oldowner;
157
158 /* Be sure to associate the file with the BufFile's resource owner */
159 oldowner = CurrentResourceOwner;
160 CurrentResourceOwner = file->resowner;
161
162 if (file->fileset == NULL)
163 pfile = OpenTemporaryFile(file->isInterXact);
164 else
165 pfile = MakeNewSharedSegment(file, file->numFiles);
166
167 Assert(pfile >= 0);
168
169 CurrentResourceOwner = oldowner;
170
171 file->files = (File *) repalloc(file->files,
172 (file->numFiles + 1) * sizeof(File));
173 file->offsets = (off_t *) repalloc(file->offsets,
174 (file->numFiles + 1) * sizeof(off_t));
175 file->files[file->numFiles] = pfile;
176 file->offsets[file->numFiles] = 0L;
177 file->numFiles++;
178 }
179
180 /*
181 * Create a BufFile for a new temporary file (which will expand to become
182 * multiple temporary files if more than MAX_PHYSICAL_FILESIZE bytes are
183 * written to it).
184 *
185 * If interXact is true, the temp file will not be automatically deleted
186 * at end of transaction.
187 *
188 * Note: if interXact is true, the caller had better be calling us in a
189 * memory context, and with a resource owner, that will survive across
190 * transaction boundaries.
191 */
192 BufFile *
BufFileCreateTemp(bool interXact)193 BufFileCreateTemp(bool interXact)
194 {
195 BufFile *file;
196 File pfile;
197
198 pfile = OpenTemporaryFile(interXact);
199 Assert(pfile >= 0);
200
201 file = makeBufFile(pfile);
202 file->isInterXact = interXact;
203
204 return file;
205 }
206
207 /*
208 * Build the name for a given segment of a given BufFile.
209 */
210 static void
SharedSegmentName(char * name,const char * buffile_name,int segment)211 SharedSegmentName(char *name, const char *buffile_name, int segment)
212 {
213 snprintf(name, MAXPGPATH, "%s.%d", buffile_name, segment);
214 }
215
216 /*
217 * Create a new segment file backing a shared BufFile.
218 */
219 static File
MakeNewSharedSegment(BufFile * buffile,int segment)220 MakeNewSharedSegment(BufFile *buffile, int segment)
221 {
222 char name[MAXPGPATH];
223 File file;
224
225 /*
226 * It is possible that there are files left over from before a crash
227 * restart with the same name. In order for BufFileOpenShared() not to
228 * get confused about how many segments there are, we'll unlink the next
229 * segment number if it already exists.
230 */
231 SharedSegmentName(name, buffile->name, segment + 1);
232 SharedFileSetDelete(buffile->fileset, name, true);
233
234 /* Create the new segment. */
235 SharedSegmentName(name, buffile->name, segment);
236 file = SharedFileSetCreate(buffile->fileset, name);
237
238 /* SharedFileSetCreate would've errored out */
239 Assert(file > 0);
240
241 return file;
242 }
243
244 /*
245 * Create a BufFile that can be discovered and opened read-only by other
246 * backends that are attached to the same SharedFileSet using the same name.
247 *
248 * The naming scheme for shared BufFiles is left up to the calling code. The
249 * name will appear as part of one or more filenames on disk, and might
250 * provide clues to administrators about which subsystem is generating
251 * temporary file data. Since each SharedFileSet object is backed by one or
252 * more uniquely named temporary directory, names don't conflict with
253 * unrelated SharedFileSet objects.
254 */
255 BufFile *
BufFileCreateShared(SharedFileSet * fileset,const char * name)256 BufFileCreateShared(SharedFileSet *fileset, const char *name)
257 {
258 BufFile *file;
259
260 file = makeBufFileCommon(1);
261 file->fileset = fileset;
262 file->name = pstrdup(name);
263 file->files = (File *) palloc(sizeof(File));
264 file->files[0] = MakeNewSharedSegment(file, 0);
265 file->readOnly = false;
266
267 return file;
268 }
269
270 /*
271 * Open a file that was previously created in another backend (or this one)
272 * with BufFileCreateShared in the same SharedFileSet using the same name.
273 * The backend that created the file must have called BufFileClose() or
274 * BufFileExportShared() to make sure that it is ready to be opened by other
275 * backends and render it read-only.
276 */
277 BufFile *
BufFileOpenShared(SharedFileSet * fileset,const char * name)278 BufFileOpenShared(SharedFileSet *fileset, const char *name)
279 {
280 BufFile *file;
281 char segment_name[MAXPGPATH];
282 Size capacity = 16;
283 File *files;
284 int nfiles = 0;
285
286 files = palloc(sizeof(File) * capacity);
287
288 /*
289 * We don't know how many segments there are, so we'll probe the
290 * filesystem to find out.
291 */
292 for (;;)
293 {
294 /* See if we need to expand our file segment array. */
295 if (nfiles + 1 > capacity)
296 {
297 capacity *= 2;
298 files = repalloc(files, sizeof(File) * capacity);
299 }
300 /* Try to load a segment. */
301 SharedSegmentName(segment_name, name, nfiles);
302 files[nfiles] = SharedFileSetOpen(fileset, segment_name);
303 if (files[nfiles] <= 0)
304 break;
305 ++nfiles;
306
307 CHECK_FOR_INTERRUPTS();
308 }
309
310 /*
311 * If we didn't find any files at all, then no BufFile exists with this
312 * name.
313 */
314 if (nfiles == 0)
315 ereport(ERROR,
316 (errcode_for_file_access(),
317 errmsg("could not open temporary file \"%s\" from BufFile \"%s\": %m",
318 segment_name, name)));
319
320 file = makeBufFileCommon(nfiles);
321 file->files = files;
322 file->readOnly = true; /* Can't write to files opened this way */
323 file->fileset = fileset;
324 file->name = pstrdup(name);
325
326 return file;
327 }
328
329 /*
330 * Delete a BufFile that was created by BufFileCreateShared in the given
331 * SharedFileSet using the given name.
332 *
333 * It is not necessary to delete files explicitly with this function. It is
334 * provided only as a way to delete files proactively, rather than waiting for
335 * the SharedFileSet to be cleaned up.
336 *
337 * Only one backend should attempt to delete a given name, and should know
338 * that it exists and has been exported or closed.
339 */
340 void
BufFileDeleteShared(SharedFileSet * fileset,const char * name)341 BufFileDeleteShared(SharedFileSet *fileset, const char *name)
342 {
343 char segment_name[MAXPGPATH];
344 int segment = 0;
345 bool found = false;
346
347 /*
348 * We don't know how many segments the file has. We'll keep deleting
349 * until we run out. If we don't manage to find even an initial segment,
350 * raise an error.
351 */
352 for (;;)
353 {
354 SharedSegmentName(segment_name, name, segment);
355 if (!SharedFileSetDelete(fileset, segment_name, true))
356 break;
357 found = true;
358 ++segment;
359
360 CHECK_FOR_INTERRUPTS();
361 }
362
363 if (!found)
364 elog(ERROR, "could not delete unknown shared BufFile \"%s\"", name);
365 }
366
367 /*
368 * BufFileExportShared --- flush and make read-only, in preparation for sharing.
369 */
370 void
BufFileExportShared(BufFile * file)371 BufFileExportShared(BufFile *file)
372 {
373 /* Must be a file belonging to a SharedFileSet. */
374 Assert(file->fileset != NULL);
375
376 /* It's probably a bug if someone calls this twice. */
377 Assert(!file->readOnly);
378
379 BufFileFlush(file);
380 file->readOnly = true;
381 }
382
383 /*
384 * Close a BufFile
385 *
386 * Like fclose(), this also implicitly FileCloses the underlying File.
387 */
388 void
BufFileClose(BufFile * file)389 BufFileClose(BufFile *file)
390 {
391 int i;
392
393 /* flush any unwritten data */
394 BufFileFlush(file);
395 /* close and delete the underlying file(s) */
396 for (i = 0; i < file->numFiles; i++)
397 FileClose(file->files[i]);
398 /* release the buffer space */
399 pfree(file->files);
400 pfree(file->offsets);
401 pfree(file);
402 }
403
404 /*
405 * BufFileLoadBuffer
406 *
407 * Load some data into buffer, if possible, starting from curOffset.
408 * At call, must have dirty = false, pos and nbytes = 0.
409 * On exit, nbytes is number of bytes loaded.
410 */
411 static void
BufFileLoadBuffer(BufFile * file)412 BufFileLoadBuffer(BufFile *file)
413 {
414 File thisfile;
415
416 /*
417 * Advance to next component file if necessary and possible.
418 */
419 if (file->curOffset >= MAX_PHYSICAL_FILESIZE &&
420 file->curFile + 1 < file->numFiles)
421 {
422 file->curFile++;
423 file->curOffset = 0L;
424 }
425
426 /*
427 * May need to reposition physical file.
428 */
429 thisfile = file->files[file->curFile];
430 if (file->curOffset != file->offsets[file->curFile])
431 {
432 if (FileSeek(thisfile, file->curOffset, SEEK_SET) != file->curOffset)
433 ereport(ERROR,
434 (errcode_for_file_access(),
435 errmsg("could not seek in file \"%s\": %m",
436 FilePathName(thisfile))));
437 file->offsets[file->curFile] = file->curOffset;
438 }
439
440 /*
441 * Read whatever we can get, up to a full bufferload.
442 */
443 file->nbytes = FileRead(thisfile,
444 file->buffer.data,
445 sizeof(file->buffer),
446 WAIT_EVENT_BUFFILE_READ);
447 if (file->nbytes < 0)
448 {
449 file->nbytes = 0;
450 ereport(ERROR,
451 (errcode_for_file_access(),
452 errmsg("could not read file \"%s\": %m",
453 FilePathName(thisfile))));
454 }
455
456 file->offsets[file->curFile] += file->nbytes;
457 /* we choose not to advance curOffset here */
458
459 if (file->nbytes > 0)
460 pgBufferUsage.temp_blks_read++;
461 }
462
463 /*
464 * BufFileDumpBuffer
465 *
466 * Dump buffer contents starting at curOffset.
467 * At call, should have dirty = true, nbytes > 0.
468 * On exit, dirty is cleared if successful write, and curOffset is advanced.
469 */
470 static void
BufFileDumpBuffer(BufFile * file)471 BufFileDumpBuffer(BufFile *file)
472 {
473 int wpos = 0;
474 int bytestowrite;
475 File thisfile;
476
477 /*
478 * Unlike BufFileLoadBuffer, we must dump the whole buffer even if it
479 * crosses a component-file boundary; so we need a loop.
480 */
481 while (wpos < file->nbytes)
482 {
483 off_t availbytes;
484
485 /*
486 * Advance to next component file if necessary and possible.
487 */
488 if (file->curOffset >= MAX_PHYSICAL_FILESIZE)
489 {
490 while (file->curFile + 1 >= file->numFiles)
491 extendBufFile(file);
492 file->curFile++;
493 file->curOffset = 0L;
494 }
495
496 /*
497 * Determine how much we need to write into this file.
498 */
499 bytestowrite = file->nbytes - wpos;
500 availbytes = MAX_PHYSICAL_FILESIZE - file->curOffset;
501
502 if ((off_t) bytestowrite > availbytes)
503 bytestowrite = (int) availbytes;
504
505 /*
506 * May need to reposition physical file.
507 */
508 thisfile = file->files[file->curFile];
509 if (file->curOffset != file->offsets[file->curFile])
510 {
511 if (FileSeek(thisfile, file->curOffset, SEEK_SET) != file->curOffset)
512 ereport(ERROR,
513 (errcode_for_file_access(),
514 errmsg("could not seek in file \"%s\": %m",
515 FilePathName(thisfile))));
516 file->offsets[file->curFile] = file->curOffset;
517 }
518 bytestowrite = FileWrite(thisfile,
519 file->buffer.data + wpos,
520 bytestowrite,
521 WAIT_EVENT_BUFFILE_WRITE);
522 if (bytestowrite <= 0)
523 ereport(ERROR,
524 (errcode_for_file_access(),
525 errmsg("could not write to file \"%s\": %m",
526 FilePathName(thisfile))));
527
528 file->offsets[file->curFile] += bytestowrite;
529 file->curOffset += bytestowrite;
530 wpos += bytestowrite;
531
532 pgBufferUsage.temp_blks_written++;
533 }
534 file->dirty = false;
535
536 /*
537 * At this point, curOffset has been advanced to the end of the buffer,
538 * ie, its original value + nbytes. We need to make it point to the
539 * logical file position, ie, original value + pos, in case that is less
540 * (as could happen due to a small backwards seek in a dirty buffer!)
541 */
542 file->curOffset -= (file->nbytes - file->pos);
543 if (file->curOffset < 0) /* handle possible segment crossing */
544 {
545 file->curFile--;
546 Assert(file->curFile >= 0);
547 file->curOffset += MAX_PHYSICAL_FILESIZE;
548 }
549
550 /*
551 * Now we can set the buffer empty without changing the logical position
552 */
553 file->pos = 0;
554 file->nbytes = 0;
555 }
556
557 /*
558 * BufFileRead
559 *
560 * Like fread() except we assume 1-byte element size and report I/O errors via
561 * ereport().
562 */
563 size_t
BufFileRead(BufFile * file,void * ptr,size_t size)564 BufFileRead(BufFile *file, void *ptr, size_t size)
565 {
566 size_t nread = 0;
567 size_t nthistime;
568
569 BufFileFlush(file);
570
571 while (size > 0)
572 {
573 if (file->pos >= file->nbytes)
574 {
575 /* Try to load more data into buffer. */
576 file->curOffset += file->pos;
577 file->pos = 0;
578 file->nbytes = 0;
579 BufFileLoadBuffer(file);
580 if (file->nbytes <= 0)
581 break; /* no more data available */
582 }
583
584 nthistime = file->nbytes - file->pos;
585 if (nthistime > size)
586 nthistime = size;
587 Assert(nthistime > 0);
588
589 memcpy(ptr, file->buffer.data + file->pos, nthistime);
590
591 file->pos += nthistime;
592 ptr = (void *) ((char *) ptr + nthistime);
593 size -= nthistime;
594 nread += nthistime;
595 }
596
597 return nread;
598 }
599
600 /*
601 * BufFileWrite
602 *
603 * Like fwrite() except we assume 1-byte element size and report errors via
604 * ereport().
605 */
606 size_t
BufFileWrite(BufFile * file,void * ptr,size_t size)607 BufFileWrite(BufFile *file, void *ptr, size_t size)
608 {
609 size_t nwritten = 0;
610 size_t nthistime;
611
612 Assert(!file->readOnly);
613
614 while (size > 0)
615 {
616 if (file->pos >= BLCKSZ)
617 {
618 /* Buffer full, dump it out */
619 if (file->dirty)
620 BufFileDumpBuffer(file);
621 else
622 {
623 /* Hmm, went directly from reading to writing? */
624 file->curOffset += file->pos;
625 file->pos = 0;
626 file->nbytes = 0;
627 }
628 }
629
630 nthistime = BLCKSZ - file->pos;
631 if (nthistime > size)
632 nthistime = size;
633 Assert(nthistime > 0);
634
635 memcpy(file->buffer.data + file->pos, ptr, nthistime);
636
637 file->dirty = true;
638 file->pos += nthistime;
639 if (file->nbytes < file->pos)
640 file->nbytes = file->pos;
641 ptr = (void *) ((char *) ptr + nthistime);
642 size -= nthistime;
643 nwritten += nthistime;
644 }
645
646 return nwritten;
647 }
648
649 /*
650 * BufFileFlush
651 *
652 * Like fflush(), except that I/O errors are reported with ereport().
653 */
654 static void
BufFileFlush(BufFile * file)655 BufFileFlush(BufFile *file)
656 {
657 if (file->dirty)
658 BufFileDumpBuffer(file);
659
660 Assert(!file->dirty);
661 }
662
663 /*
664 * BufFileSeek
665 *
666 * Like fseek(), except that target position needs two values in order to
667 * work when logical filesize exceeds maximum value representable by off_t.
668 * We do not support relative seeks across more than that, however.
669 * I/O errors are reported by ereport().
670 *
671 * Result is 0 if OK, EOF if not. Logical position is not moved if an
672 * impossible seek is attempted.
673 */
674 int
BufFileSeek(BufFile * file,int fileno,off_t offset,int whence)675 BufFileSeek(BufFile *file, int fileno, off_t offset, int whence)
676 {
677 int newFile;
678 off_t newOffset;
679
680 switch (whence)
681 {
682 case SEEK_SET:
683 if (fileno < 0)
684 return EOF;
685 newFile = fileno;
686 newOffset = offset;
687 break;
688 case SEEK_CUR:
689
690 /*
691 * Relative seek considers only the signed offset, ignoring
692 * fileno. Note that large offsets (> 1 gig) risk overflow in this
693 * add, unless we have 64-bit off_t.
694 */
695 newFile = file->curFile;
696 newOffset = (file->curOffset + file->pos) + offset;
697 break;
698 #ifdef NOT_USED
699 case SEEK_END:
700 /* could be implemented, not needed currently */
701 break;
702 #endif
703 default:
704 elog(ERROR, "invalid whence: %d", whence);
705 return EOF;
706 }
707 while (newOffset < 0)
708 {
709 if (--newFile < 0)
710 return EOF;
711 newOffset += MAX_PHYSICAL_FILESIZE;
712 }
713 if (newFile == file->curFile &&
714 newOffset >= file->curOffset &&
715 newOffset <= file->curOffset + file->nbytes)
716 {
717 /*
718 * Seek is to a point within existing buffer; we can just adjust
719 * pos-within-buffer, without flushing buffer. Note this is OK
720 * whether reading or writing, but buffer remains dirty if we were
721 * writing.
722 */
723 file->pos = (int) (newOffset - file->curOffset);
724 return 0;
725 }
726 /* Otherwise, must reposition buffer, so flush any dirty data */
727 BufFileFlush(file);
728
729 /*
730 * At this point and no sooner, check for seek past last segment. The
731 * above flush could have created a new segment, so checking sooner would
732 * not work (at least not with this code).
733 */
734
735 /* convert seek to "start of next seg" to "end of last seg" */
736 if (newFile == file->numFiles && newOffset == 0)
737 {
738 newFile--;
739 newOffset = MAX_PHYSICAL_FILESIZE;
740 }
741 while (newOffset > MAX_PHYSICAL_FILESIZE)
742 {
743 if (++newFile >= file->numFiles)
744 return EOF;
745 newOffset -= MAX_PHYSICAL_FILESIZE;
746 }
747 if (newFile >= file->numFiles)
748 return EOF;
749 /* Seek is OK! */
750 file->curFile = newFile;
751 file->curOffset = newOffset;
752 file->pos = 0;
753 file->nbytes = 0;
754 return 0;
755 }
756
757 void
BufFileTell(BufFile * file,int * fileno,off_t * offset)758 BufFileTell(BufFile *file, int *fileno, off_t *offset)
759 {
760 *fileno = file->curFile;
761 *offset = file->curOffset + file->pos;
762 }
763
764 /*
765 * BufFileSeekBlock --- block-oriented seek
766 *
767 * Performs absolute seek to the start of the n'th BLCKSZ-sized block of
768 * the file. Note that users of this interface will fail if their files
769 * exceed BLCKSZ * LONG_MAX bytes, but that is quite a lot; we don't work
770 * with tables bigger than that, either...
771 *
772 * Result is 0 if OK, EOF if not. Logical position is not moved if an
773 * impossible seek is attempted.
774 */
775 int
BufFileSeekBlock(BufFile * file,long blknum)776 BufFileSeekBlock(BufFile *file, long blknum)
777 {
778 return BufFileSeek(file,
779 (int) (blknum / BUFFILE_SEG_SIZE),
780 (off_t) (blknum % BUFFILE_SEG_SIZE) * BLCKSZ,
781 SEEK_SET);
782 }
783
784 #ifdef NOT_USED
785 /*
786 * BufFileTellBlock --- block-oriented tell
787 *
788 * Any fractional part of a block in the current seek position is ignored.
789 */
790 long
BufFileTellBlock(BufFile * file)791 BufFileTellBlock(BufFile *file)
792 {
793 long blknum;
794
795 blknum = (file->curOffset + file->pos) / BLCKSZ;
796 blknum += file->curFile * BUFFILE_SEG_SIZE;
797 return blknum;
798 }
799
800 #endif
801
802 /*
803 * Return the current shared BufFile size.
804 *
805 * Counts any holes left behind by BufFileAppend as part of the size.
806 * ereport()s on failure.
807 */
808 int64
BufFileSize(BufFile * file)809 BufFileSize(BufFile *file)
810 {
811 int64 lastFileSize;
812
813 Assert(file->fileset != NULL);
814
815 /* Get the size of the last physical file by seeking to end. */
816 lastFileSize = FileSeek(file->files[file->numFiles - 1], 0, SEEK_END);
817 if (lastFileSize < 0)
818 ereport(ERROR,
819 (errcode_for_file_access(),
820 errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
821 FilePathName(file->files[file->numFiles - 1]),
822 file->name)));
823 file->offsets[file->numFiles - 1] = lastFileSize;
824
825 return ((file->numFiles - 1) * (int64) MAX_PHYSICAL_FILESIZE) +
826 lastFileSize;
827 }
828
829 /*
830 * Append the contents of source file (managed within shared fileset) to
831 * end of target file (managed within same shared fileset).
832 *
833 * Note that operation subsumes ownership of underlying resources from
834 * "source". Caller should never call BufFileClose against source having
835 * called here first. Resource owners for source and target must match,
836 * too.
837 *
838 * This operation works by manipulating lists of segment files, so the
839 * file content is always appended at a MAX_PHYSICAL_FILESIZE-aligned
840 * boundary, typically creating empty holes before the boundary. These
841 * areas do not contain any interesting data, and cannot be read from by
842 * caller.
843 *
844 * Returns the block number within target where the contents of source
845 * begins. Caller should apply this as an offset when working off block
846 * positions that are in terms of the original BufFile space.
847 */
848 long
BufFileAppend(BufFile * target,BufFile * source)849 BufFileAppend(BufFile *target, BufFile *source)
850 {
851 long startBlock = target->numFiles * BUFFILE_SEG_SIZE;
852 int newNumFiles = target->numFiles + source->numFiles;
853 int i;
854
855 Assert(target->fileset != NULL);
856 Assert(source->readOnly);
857 Assert(!source->dirty);
858 Assert(source->fileset != NULL);
859
860 if (target->resowner != source->resowner)
861 elog(ERROR, "could not append BufFile with non-matching resource owner");
862
863 target->files = (File *)
864 repalloc(target->files, sizeof(File) * newNumFiles);
865 target->offsets = (off_t *)
866 repalloc(target->offsets, sizeof(off_t) * newNumFiles);
867 for (i = target->numFiles; i < newNumFiles; i++)
868 {
869 target->files[i] = source->files[i - target->numFiles];
870 target->offsets[i] = source->offsets[i - target->numFiles];
871 }
872 target->numFiles = newNumFiles;
873
874 return startBlock;
875 }
876