Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * buffile.c
4 : : * Management of large buffered temporary files.
5 : : *
6 : : * Portions Copyright (c) 1996-2026, 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/buffile.h"
53 : : #include "storage/bufmgr.h"
54 : : #include "storage/fd.h"
55 : : #include "utils/resowner.h"
56 : : #include "utils/wait_event.h"
57 : :
58 : : /*
59 : : * We break BufFiles into gigabyte-sized segments, regardless of RELSEG_SIZE.
60 : : * The reason is that we'd like large BufFiles to be spread across multiple
61 : : * tablespaces when available.
62 : : */
63 : : #define MAX_PHYSICAL_FILESIZE 0x40000000
64 : : #define BUFFILE_SEG_SIZE (MAX_PHYSICAL_FILESIZE / BLCKSZ)
65 : :
66 : : /*
67 : : * This data structure represents a buffered file that consists of one or
68 : : * more physical files (each accessed through a virtual file descriptor
69 : : * managed by fd.c).
70 : : */
71 : : struct BufFile
72 : : {
73 : : int numFiles; /* number of physical files in set */
74 : : /* all files except the last have length exactly MAX_PHYSICAL_FILESIZE */
75 : : File *files; /* palloc'd array with numFiles entries */
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 : : FileSet *fileset; /* space for fileset based segment files */
82 : : const char *name; /* name of fileset based BufFile */
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 : : pgoff_t curOffset; /* offset part of current pos */
97 : : int64 pos; /* next read/write position in buffer */
98 : : int64 nbytes; /* total # of valid bytes in buffer */
99 : :
100 : : /*
101 : : * XXX Should ideally use PGIOAlignedBlock, but might need a way to avoid
102 : : * wasting per-file alignment padding when some users create many files.
103 : : */
104 : : PGAlignedBlock buffer;
105 : : };
106 : :
107 : : static BufFile *makeBufFileCommon(int nfiles);
108 : : static BufFile *makeBufFile(File firstfile);
109 : : static void extendBufFile(BufFile *file);
110 : : static void BufFileLoadBuffer(BufFile *file);
111 : : static void BufFileDumpBuffer(BufFile *file);
112 : : static void BufFileFlush(BufFile *file);
113 : : static File MakeNewFileSetSegment(BufFile *buffile, int segment);
114 : :
115 : : /*
116 : : * Create BufFile and perform the common initialization.
117 : : */
118 : : static BufFile *
119 : 6511 : makeBufFileCommon(int nfiles)
120 : : {
121 : 6511 : BufFile *file = palloc_object(BufFile);
122 : :
123 : 6511 : file->numFiles = nfiles;
124 : 6511 : file->isInterXact = false;
125 : 6511 : file->dirty = false;
126 : 6511 : file->resowner = CurrentResourceOwner;
127 : 6511 : file->curFile = 0;
128 : 6511 : file->curOffset = 0;
129 : 6511 : file->pos = 0;
130 : 6511 : file->nbytes = 0;
131 : :
132 : 6511 : return file;
133 : : }
134 : :
135 : : /*
136 : : * Create a BufFile given the first underlying physical file.
137 : : * NOTE: caller must set isInterXact if appropriate.
138 : : */
139 : : static BufFile *
140 : 2051 : makeBufFile(File firstfile)
141 : : {
142 : 2051 : BufFile *file = makeBufFileCommon(1);
143 : :
144 : 2051 : file->files = palloc_object(File);
145 : 2051 : file->files[0] = firstfile;
146 : 2051 : file->readOnly = false;
147 : 2051 : file->fileset = NULL;
148 : 2051 : file->name = NULL;
149 : :
150 : 2051 : return file;
151 : : }
152 : :
153 : : /*
154 : : * Add another component temp file.
155 : : */
156 : : static void
157 : 0 : extendBufFile(BufFile *file)
158 : : {
159 : : File pfile;
160 : : ResourceOwner oldowner;
161 : :
162 : : /* Be sure to associate the file with the BufFile's resource owner */
163 : 0 : oldowner = CurrentResourceOwner;
164 : 0 : CurrentResourceOwner = file->resowner;
165 : :
166 [ # # ]: 0 : if (file->fileset == NULL)
167 : 0 : pfile = OpenTemporaryFile(file->isInterXact);
168 : : else
169 : 0 : pfile = MakeNewFileSetSegment(file, file->numFiles);
170 : :
171 : : Assert(pfile >= 0);
172 : :
173 : 0 : CurrentResourceOwner = oldowner;
174 : :
175 : 0 : file->files = repalloc_array(file->files, File, file->numFiles + 1);
176 : 0 : file->files[file->numFiles] = pfile;
177 : 0 : file->numFiles++;
178 : 0 : }
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 *
193 : 2051 : BufFileCreateTemp(bool interXact)
194 : : {
195 : : BufFile *file;
196 : : File pfile;
197 : :
198 : : /*
199 : : * Ensure that temp tablespaces are set up for OpenTemporaryFile to use.
200 : : * Possibly the caller will have done this already, but it seems useful to
201 : : * double-check here. Failure to do this at all would result in the temp
202 : : * files always getting placed in the default tablespace, which is a
203 : : * pretty hard-to-detect bug. Callers may prefer to do it earlier if they
204 : : * want to be sure that any required catalog access is done in some other
205 : : * resource context.
206 : : */
207 : 2051 : PrepareTempTablespaces();
208 : :
209 : 2051 : pfile = OpenTemporaryFile(interXact);
210 : : Assert(pfile >= 0);
211 : :
212 : 2051 : file = makeBufFile(pfile);
213 : 2051 : file->isInterXact = interXact;
214 : :
215 : 2051 : return file;
216 : : }
217 : :
218 : : /*
219 : : * Build the name for a given segment of a given BufFile.
220 : : */
221 : : static void
222 : 9584 : FileSetSegmentName(char *name, const char *buffile_name, int segment)
223 : : {
224 : 9584 : snprintf(name, MAXPGPATH, "%s.%d", buffile_name, segment);
225 : 9584 : }
226 : :
227 : : /*
228 : : * Create a new segment file backing a fileset based BufFile.
229 : : */
230 : : static File
231 : 2014 : MakeNewFileSetSegment(BufFile *buffile, int segment)
232 : : {
233 : : char name[MAXPGPATH];
234 : : File file;
235 : :
236 : : /*
237 : : * It is possible that there are files left over from before a crash
238 : : * restart with the same name. In order for BufFileOpenFileSet() not to
239 : : * get confused about how many segments there are, we'll unlink the next
240 : : * segment number if it already exists.
241 : : */
242 : 2014 : FileSetSegmentName(name, buffile->name, segment + 1);
243 : 2014 : FileSetDelete(buffile->fileset, name, true);
244 : :
245 : : /* Create the new segment. */
246 : 2014 : FileSetSegmentName(name, buffile->name, segment);
247 : 2014 : file = FileSetCreate(buffile->fileset, name);
248 : :
249 : : /* FileSetCreate would've errored out */
250 : : Assert(file > 0);
251 : :
252 : 2014 : return file;
253 : : }
254 : :
255 : : /*
256 : : * Create a BufFile that can be discovered and opened read-only by other
257 : : * backends that are attached to the same SharedFileSet using the same name.
258 : : *
259 : : * The naming scheme for fileset based BufFiles is left up to the calling code.
260 : : * The name will appear as part of one or more filenames on disk, and might
261 : : * provide clues to administrators about which subsystem is generating
262 : : * temporary file data. Since each SharedFileSet object is backed by one or
263 : : * more uniquely named temporary directory, names don't conflict with
264 : : * unrelated SharedFileSet objects.
265 : : */
266 : : BufFile *
267 : 2014 : BufFileCreateFileSet(FileSet *fileset, const char *name)
268 : : {
269 : : BufFile *file;
270 : :
271 : 2014 : file = makeBufFileCommon(1);
272 : 2014 : file->fileset = fileset;
273 : 2014 : file->name = pstrdup(name);
274 : 2014 : file->files = palloc_object(File);
275 : 2014 : file->files[0] = MakeNewFileSetSegment(file, 0);
276 : 2014 : file->readOnly = false;
277 : :
278 : 2014 : return file;
279 : : }
280 : :
281 : : /*
282 : : * Open a file that was previously created in another backend (or this one)
283 : : * with BufFileCreateFileSet in the same FileSet using the same name.
284 : : * The backend that created the file must have called BufFileClose() or
285 : : * BufFileExportFileSet() to make sure that it is ready to be opened by other
286 : : * backends and render it read-only. If missing_ok is true, which indicates
287 : : * that missing files can be safely ignored, then return NULL if the BufFile
288 : : * with the given name is not found, otherwise, throw an error.
289 : : */
290 : : BufFile *
291 : 2719 : BufFileOpenFileSet(FileSet *fileset, const char *name, int mode,
292 : : bool missing_ok)
293 : : {
294 : : BufFile *file;
295 : : char segment_name[MAXPGPATH];
296 : 2719 : Size capacity = 16;
297 : : File *files;
298 : 2719 : int nfiles = 0;
299 : :
300 : 2719 : files = palloc_array(File, capacity);
301 : :
302 : : /*
303 : : * We don't know how many segments there are, so we'll probe the
304 : : * filesystem to find out.
305 : : */
306 : : for (;;)
307 : : {
308 : : /* See if we need to expand our file segment array. */
309 [ - + ]: 5165 : if (nfiles + 1 > capacity)
310 : : {
311 : 0 : capacity *= 2;
312 : 0 : files = repalloc_array(files, File, capacity);
313 : : }
314 : : /* Try to load a segment. */
315 : 5165 : FileSetSegmentName(segment_name, name, nfiles);
316 : 5165 : files[nfiles] = FileSetOpen(fileset, segment_name, mode);
317 [ + + ]: 5165 : if (files[nfiles] <= 0)
318 : 2719 : break;
319 : 2446 : ++nfiles;
320 : :
321 [ + + ]: 2446 : CHECK_FOR_INTERRUPTS();
322 : : }
323 : :
324 : : /*
325 : : * If we didn't find any files at all, then no BufFile exists with this
326 : : * name.
327 : : */
328 [ + + ]: 2719 : if (nfiles == 0)
329 : : {
330 : : /* free the memory */
331 : 273 : pfree(files);
332 : :
333 [ + - ]: 273 : if (missing_ok)
334 : 273 : return NULL;
335 : :
336 [ # # ]: 0 : ereport(ERROR,
337 : : (errcode_for_file_access(),
338 : : errmsg("could not open temporary file \"%s\" from BufFile \"%s\": %m",
339 : : segment_name, name)));
340 : : }
341 : :
342 : 2446 : file = makeBufFileCommon(nfiles);
343 : 2446 : file->files = files;
344 : 2446 : file->readOnly = (mode == O_RDONLY);
345 : 2446 : file->fileset = fileset;
346 : 2446 : file->name = pstrdup(name);
347 : :
348 : 2446 : return file;
349 : : }
350 : :
351 : : /*
352 : : * Delete a BufFile that was created by BufFileCreateFileSet in the given
353 : : * FileSet using the given name.
354 : : *
355 : : * It is not necessary to delete files explicitly with this function. It is
356 : : * provided only as a way to delete files proactively, rather than waiting for
357 : : * the FileSet to be cleaned up.
358 : : *
359 : : * Only one backend should attempt to delete a given name, and should know
360 : : * that it exists and has been exported or closed otherwise missing_ok should
361 : : * be passed true.
362 : : */
363 : : void
364 : 352 : BufFileDeleteFileSet(FileSet *fileset, const char *name, bool missing_ok)
365 : : {
366 : : char segment_name[MAXPGPATH];
367 : 352 : int segment = 0;
368 : 352 : bool found = false;
369 : :
370 : : /*
371 : : * We don't know how many segments the file has. We'll keep deleting
372 : : * until we run out. If we don't manage to find even an initial segment,
373 : : * raise an error.
374 : : */
375 : : for (;;)
376 : : {
377 : 391 : FileSetSegmentName(segment_name, name, segment);
378 [ + + ]: 391 : if (!FileSetDelete(fileset, segment_name, true))
379 : 352 : break;
380 : 39 : found = true;
381 : 39 : ++segment;
382 : :
383 [ - + ]: 39 : CHECK_FOR_INTERRUPTS();
384 : : }
385 : :
386 [ + + - + ]: 352 : if (!found && !missing_ok)
387 [ # # ]: 0 : elog(ERROR, "could not delete unknown BufFile \"%s\"", name);
388 : 352 : }
389 : :
390 : : /*
391 : : * BufFileExportFileSet --- flush and make read-only, in preparation for sharing.
392 : : */
393 : : void
394 : 379 : BufFileExportFileSet(BufFile *file)
395 : : {
396 : : /* Must be a file belonging to a FileSet. */
397 : : Assert(file->fileset != NULL);
398 : :
399 : : /* It's probably a bug if someone calls this twice. */
400 : : Assert(!file->readOnly);
401 : :
402 : 379 : BufFileFlush(file);
403 : 379 : file->readOnly = true;
404 : 379 : }
405 : :
406 : : /*
407 : : * Close a BufFile
408 : : *
409 : : * Like fclose(), this also implicitly FileCloses the underlying File.
410 : : */
411 : : void
412 : 6350 : BufFileClose(BufFile *file)
413 : : {
414 : : int i;
415 : :
416 : : /* flush any unwritten data */
417 : 6350 : BufFileFlush(file);
418 : : /* close and delete the underlying file(s) */
419 [ + + ]: 12853 : for (i = 0; i < file->numFiles; i++)
420 : 6503 : FileClose(file->files[i]);
421 : : /* release the buffer space */
422 : 6350 : pfree(file->files);
423 : 6350 : pfree(file);
424 : 6350 : }
425 : :
426 : : /*
427 : : * BufFileLoadBuffer
428 : : *
429 : : * Load some data into buffer, if possible, starting from curOffset.
430 : : * At call, must have dirty = false, pos and nbytes = 0.
431 : : * On exit, nbytes is number of bytes loaded.
432 : : */
433 : : static void
434 : 64309 : BufFileLoadBuffer(BufFile *file)
435 : : {
436 : : File thisfile;
437 : : instr_time io_start;
438 : : instr_time io_time;
439 : : ssize_t rc;
440 : :
441 : : /*
442 : : * Advance to next component file if necessary and possible.
443 : : */
444 [ - + ]: 64309 : if (file->curOffset >= MAX_PHYSICAL_FILESIZE &&
445 [ # # ]: 0 : file->curFile + 1 < file->numFiles)
446 : : {
447 : 0 : file->curFile++;
448 : 0 : file->curOffset = 0;
449 : : }
450 : :
451 : 64309 : thisfile = file->files[file->curFile];
452 : :
453 [ - + ]: 64309 : if (track_io_timing)
454 : 0 : INSTR_TIME_SET_CURRENT(io_start);
455 : : else
456 : 64309 : INSTR_TIME_SET_ZERO(io_start);
457 : :
458 : : /*
459 : : * Read whatever we can get, up to a full bufferload.
460 : : */
461 : 64309 : rc = FileRead(thisfile,
462 : 64309 : file->buffer.data,
463 : : sizeof(file->buffer.data),
464 : : file->curOffset,
465 : : WAIT_EVENT_BUFFILE_READ);
466 [ - + ]: 64309 : if (rc < 0)
467 : : {
468 : 0 : file->nbytes = 0;
469 [ # # ]: 0 : ereport(ERROR,
470 : : (errcode_for_file_access(),
471 : : errmsg("could not read file \"%s\": %m",
472 : : FilePathName(thisfile))));
473 : : }
474 : :
475 : 64309 : file->nbytes = rc;
476 : :
477 [ - + ]: 64309 : if (track_io_timing)
478 : : {
479 : 0 : INSTR_TIME_SET_CURRENT(io_time);
480 : 0 : INSTR_TIME_ACCUM_DIFF(pgBufferUsage.temp_blk_read_time, io_time, io_start);
481 : : }
482 : :
483 : : /* we choose not to advance curOffset here */
484 : :
485 [ + + ]: 64309 : if (file->nbytes > 0)
486 : 62543 : pgBufferUsage.temp_blks_read++;
487 : 64309 : }
488 : :
489 : : /*
490 : : * BufFileDumpBuffer
491 : : *
492 : : * Dump buffer contents starting at curOffset.
493 : : * At call, should have dirty = true, nbytes > 0.
494 : : * On exit, dirty is cleared if successful write, and curOffset is advanced.
495 : : */
496 : : static void
497 : 69532 : BufFileDumpBuffer(BufFile *file)
498 : : {
499 : 69532 : int64 wpos = 0;
500 : : File thisfile;
501 : :
502 : : /*
503 : : * Unlike BufFileLoadBuffer, we must dump the whole buffer even if it
504 : : * crosses a component-file boundary; so we need a loop.
505 : : */
506 [ + + ]: 139064 : while (wpos < file->nbytes)
507 : : {
508 : : int64 availbytes;
509 : : instr_time io_start;
510 : : instr_time io_time;
511 : : size_t bytestowrite;
512 : : ssize_t rc;
513 : :
514 : : /*
515 : : * Advance to next component file if necessary and possible.
516 : : */
517 [ - + ]: 69532 : if (file->curOffset >= MAX_PHYSICAL_FILESIZE)
518 : : {
519 [ # # ]: 0 : while (file->curFile + 1 >= file->numFiles)
520 : 0 : extendBufFile(file);
521 : 0 : file->curFile++;
522 : 0 : file->curOffset = 0;
523 : : }
524 : :
525 : : /*
526 : : * Determine how much we need to write into this file.
527 : : */
528 : 69532 : bytestowrite = file->nbytes - wpos;
529 : 69532 : availbytes = MAX_PHYSICAL_FILESIZE - file->curOffset;
530 : :
531 [ - + ]: 69532 : if (bytestowrite > availbytes)
532 : 0 : bytestowrite = availbytes;
533 : :
534 : 69532 : thisfile = file->files[file->curFile];
535 : :
536 [ - + ]: 69532 : if (track_io_timing)
537 : 0 : INSTR_TIME_SET_CURRENT(io_start);
538 : : else
539 : 69532 : INSTR_TIME_SET_ZERO(io_start);
540 : :
541 : 69532 : rc = FileWrite(thisfile,
542 : 69532 : file->buffer.data + wpos,
543 : : bytestowrite,
544 : : file->curOffset,
545 : : WAIT_EVENT_BUFFILE_WRITE);
546 [ - + ]: 69532 : if (rc <= 0)
547 [ # # ]: 0 : ereport(ERROR,
548 : : (errcode_for_file_access(),
549 : : errmsg("could not write to file \"%s\": %m",
550 : : FilePathName(thisfile))));
551 : :
552 [ - + ]: 69532 : if (track_io_timing)
553 : : {
554 : 0 : INSTR_TIME_SET_CURRENT(io_time);
555 : 0 : INSTR_TIME_ACCUM_DIFF(pgBufferUsage.temp_blk_write_time, io_time, io_start);
556 : : }
557 : :
558 : 69532 : file->curOffset += rc;
559 : 69532 : wpos += rc;
560 : :
561 : 69532 : pgBufferUsage.temp_blks_written++;
562 : : }
563 : 69532 : file->dirty = false;
564 : :
565 : : /*
566 : : * At this point, curOffset has been advanced to the end of the buffer,
567 : : * ie, its original value + nbytes. We need to make it point to the
568 : : * logical file position, ie, original value + pos, in case that is less
569 : : * (as could happen due to a small backwards seek in a dirty buffer!)
570 : : */
571 : 69532 : file->curOffset -= (file->nbytes - file->pos);
572 [ - + ]: 69532 : if (file->curOffset < 0) /* handle possible segment crossing */
573 : : {
574 : 0 : file->curFile--;
575 : : Assert(file->curFile >= 0);
576 : 0 : file->curOffset += MAX_PHYSICAL_FILESIZE;
577 : : }
578 : :
579 : : /*
580 : : * Now we can set the buffer empty without changing the logical position
581 : : */
582 : 69532 : file->pos = 0;
583 : 69532 : file->nbytes = 0;
584 : 69532 : }
585 : :
586 : : /*
587 : : * BufFileRead variants
588 : : *
589 : : * Like fread() except we assume 1-byte element size and report I/O errors via
590 : : * ereport().
591 : : *
592 : : * If 'exact' is true, then an error is also raised if the number of bytes
593 : : * read is not exactly 'size' (no short reads). If 'exact' and 'eofOK' are
594 : : * true, then reading zero bytes is ok.
595 : : */
596 : : static size_t
597 : 16209488 : BufFileReadCommon(BufFile *file, void *ptr, size_t size, bool exact, bool eofOK)
598 : : {
599 : 16209488 : size_t start_size = size;
600 : 16209488 : size_t nread = 0;
601 : : size_t nthistime;
602 : :
603 : 16209488 : BufFileFlush(file);
604 : :
605 [ + + ]: 32433712 : while (size > 0)
606 : : {
607 [ + + ]: 16225990 : if (file->pos >= file->nbytes)
608 : : {
609 : : /* Try to load more data into buffer. */
610 : 64309 : file->curOffset += file->pos;
611 : 64309 : file->pos = 0;
612 : 64309 : file->nbytes = 0;
613 : 64309 : BufFileLoadBuffer(file);
614 [ + + ]: 64309 : if (file->nbytes <= 0)
615 : 1766 : break; /* no more data available */
616 : : }
617 : :
618 : 16224224 : nthistime = file->nbytes - file->pos;
619 [ + + ]: 16224224 : if (nthistime > size)
620 : 16163746 : nthistime = size;
621 : : Assert(nthistime > 0);
622 : :
623 : 16224224 : memcpy(ptr, file->buffer.data + file->pos, nthistime);
624 : :
625 : 16224224 : file->pos += nthistime;
626 : 16224224 : ptr = (char *) ptr + nthistime;
627 : 16224224 : size -= nthistime;
628 : 16224224 : nread += nthistime;
629 : : }
630 : :
631 [ + - + + ]: 16209488 : if (exact &&
632 [ + - - + ]: 1766 : (nread != start_size && !(nread == 0 && eofOK)))
633 [ # # # # ]: 0 : ereport(ERROR,
634 : : errcode_for_file_access(),
635 : : file->name ?
636 : : errmsg("could not read from file set \"%s\": read only %zu of %zu bytes",
637 : : file->name, nread, start_size) :
638 : : errmsg("could not read from temporary file: read only %zu of %zu bytes",
639 : : nread, start_size));
640 : :
641 : 16209488 : return nread;
642 : : }
643 : :
644 : : /*
645 : : * Legacy interface where the caller needs to check for end of file or short
646 : : * reads.
647 : : */
648 : : size_t
649 : 0 : BufFileRead(BufFile *file, void *ptr, size_t size)
650 : : {
651 : 0 : return BufFileReadCommon(file, ptr, size, false, false);
652 : : }
653 : :
654 : : /*
655 : : * Require read of exactly the specified size.
656 : : */
657 : : void
658 : 10552681 : BufFileReadExact(BufFile *file, void *ptr, size_t size)
659 : : {
660 : 10552681 : BufFileReadCommon(file, ptr, size, true, false);
661 : 10552681 : }
662 : :
663 : : /*
664 : : * Require read of exactly the specified size, but optionally allow end of
665 : : * file (in which case 0 is returned).
666 : : */
667 : : size_t
668 : 5656807 : BufFileReadMaybeEOF(BufFile *file, void *ptr, size_t size, bool eofOK)
669 : : {
670 : 5656807 : return BufFileReadCommon(file, ptr, size, true, eofOK);
671 : : }
672 : :
673 : : /*
674 : : * BufFileWrite
675 : : *
676 : : * Like fwrite() except we assume 1-byte element size and report errors via
677 : : * ereport().
678 : : */
679 : : void
680 : 11752176 : BufFileWrite(BufFile *file, const void *ptr, size_t size)
681 : : {
682 : : size_t nthistime;
683 : :
684 : : Assert(!file->readOnly);
685 : :
686 [ + + ]: 23529440 : while (size > 0)
687 : : {
688 [ + + ]: 11777264 : if (file->pos >= BLCKSZ)
689 : : {
690 : : /* Buffer full, dump it out */
691 [ + + ]: 42452 : if (file->dirty)
692 : 42136 : BufFileDumpBuffer(file);
693 : : else
694 : : {
695 : : /* Hmm, went directly from reading to writing? */
696 : 316 : file->curOffset += file->pos;
697 : 316 : file->pos = 0;
698 : 316 : file->nbytes = 0;
699 : : }
700 : : }
701 : :
702 : 11777264 : nthistime = BLCKSZ - file->pos;
703 [ + + ]: 11777264 : if (nthistime > size)
704 : 11710094 : nthistime = size;
705 : : Assert(nthistime > 0);
706 : :
707 : 11777264 : memcpy(file->buffer.data + file->pos, ptr, nthistime);
708 : :
709 : 11777264 : file->dirty = true;
710 : 11777264 : file->pos += nthistime;
711 [ + + ]: 11777264 : if (file->nbytes < file->pos)
712 : 11774688 : file->nbytes = file->pos;
713 : 11777264 : ptr = (const char *) ptr + nthistime;
714 : 11777264 : size -= nthistime;
715 : : }
716 : 11752176 : }
717 : :
718 : : /*
719 : : * BufFileFlush
720 : : *
721 : : * Like fflush(), except that I/O errors are reported with ereport().
722 : : */
723 : : static void
724 : 16251866 : BufFileFlush(BufFile *file)
725 : : {
726 [ + + ]: 16251866 : if (file->dirty)
727 : 27396 : BufFileDumpBuffer(file);
728 : :
729 : : Assert(!file->dirty);
730 : 16251866 : }
731 : :
732 : : /*
733 : : * BufFileSeek
734 : : *
735 : : * Like fseek(), except that target position needs two values in order to
736 : : * work when logical filesize exceeds maximum value representable by pgoff_t.
737 : : * We do not support relative seeks across more than that, however.
738 : : * I/O errors are reported by ereport().
739 : : *
740 : : * Result is 0 if OK, EOF if not. Logical position is not moved if an
741 : : * impossible seek is attempted.
742 : : */
743 : : int
744 : 72623 : BufFileSeek(BufFile *file, int fileno, pgoff_t offset, int whence)
745 : : {
746 : : int newFile;
747 : : pgoff_t newOffset;
748 : :
749 [ + - + - ]: 72623 : switch (whence)
750 : : {
751 : 72288 : case SEEK_SET:
752 [ - + ]: 72288 : if (fileno < 0)
753 : 0 : return EOF;
754 : 72288 : newFile = fileno;
755 : 72288 : newOffset = offset;
756 : 72288 : break;
757 : 0 : case SEEK_CUR:
758 : :
759 : : /*
760 : : * Relative seek considers only the signed offset, ignoring
761 : : * fileno.
762 : : */
763 : 0 : newFile = file->curFile;
764 : 0 : newOffset = (file->curOffset + file->pos) + offset;
765 : 0 : break;
766 : 335 : case SEEK_END:
767 : :
768 : : /*
769 : : * The file size of the last file gives us the end offset of that
770 : : * file.
771 : : */
772 : 335 : newFile = file->numFiles - 1;
773 : 335 : newOffset = FileSize(file->files[file->numFiles - 1]);
774 [ - + ]: 335 : if (newOffset < 0)
775 [ # # ]: 0 : ereport(ERROR,
776 : : (errcode_for_file_access(),
777 : : errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
778 : : FilePathName(file->files[file->numFiles - 1]),
779 : : file->name)));
780 : 335 : break;
781 : 0 : default:
782 [ # # ]: 0 : elog(ERROR, "invalid whence: %d", whence);
783 : : return EOF;
784 : : }
785 [ - + ]: 72623 : while (newOffset < 0)
786 : : {
787 [ # # ]: 0 : if (--newFile < 0)
788 : 0 : return EOF;
789 : 0 : newOffset += MAX_PHYSICAL_FILESIZE;
790 : : }
791 [ + + ]: 72623 : if (newFile == file->curFile &&
792 [ + + ]: 72470 : newOffset >= file->curOffset &&
793 [ + + ]: 52967 : newOffset <= file->curOffset + file->nbytes)
794 : : {
795 : : /*
796 : : * Seek is to a point within existing buffer; we can just adjust
797 : : * pos-within-buffer, without flushing buffer. Note this is OK
798 : : * whether reading or writing, but buffer remains dirty if we were
799 : : * writing.
800 : : */
801 : 36974 : file->pos = (int64) (newOffset - file->curOffset);
802 : 36974 : return 0;
803 : : }
804 : : /* Otherwise, must reposition buffer, so flush any dirty data */
805 : 35649 : BufFileFlush(file);
806 : :
807 : : /*
808 : : * At this point and no sooner, check for seek past last segment. The
809 : : * above flush could have created a new segment, so checking sooner would
810 : : * not work (at least not with this code).
811 : : */
812 : :
813 : : /* convert seek to "start of next seg" to "end of last seg" */
814 [ - + - - ]: 35649 : if (newFile == file->numFiles && newOffset == 0)
815 : : {
816 : 0 : newFile--;
817 : 0 : newOffset = MAX_PHYSICAL_FILESIZE;
818 : : }
819 [ - + ]: 35649 : while (newOffset > MAX_PHYSICAL_FILESIZE)
820 : : {
821 [ # # ]: 0 : if (++newFile >= file->numFiles)
822 : 0 : return EOF;
823 : 0 : newOffset -= MAX_PHYSICAL_FILESIZE;
824 : : }
825 [ - + ]: 35649 : if (newFile >= file->numFiles)
826 : 0 : return EOF;
827 : : /* Seek is OK! */
828 : 35649 : file->curFile = newFile;
829 : 35649 : file->curOffset = newOffset;
830 : 35649 : file->pos = 0;
831 : 35649 : file->nbytes = 0;
832 : 35649 : return 0;
833 : : }
834 : :
835 : : void
836 : 88672 : BufFileTell(BufFile *file, int *fileno, pgoff_t *offset)
837 : : {
838 : 88672 : *fileno = file->curFile;
839 : 88672 : *offset = file->curOffset + file->pos;
840 : 88672 : }
841 : :
842 : : /*
843 : : * BufFileSeekBlock --- block-oriented seek
844 : : *
845 : : * Performs absolute seek to the start of the n'th BLCKSZ-sized block of
846 : : * the file. Note that users of this interface will fail if their files
847 : : * exceed BLCKSZ * PG_INT64_MAX bytes, but that is quite a lot; we don't
848 : : * work with tables bigger than that, either...
849 : : *
850 : : * Result is 0 if OK, EOF if not. Logical position is not moved if an
851 : : * impossible seek is attempted.
852 : : */
853 : : int
854 : 70357 : BufFileSeekBlock(BufFile *file, int64 blknum)
855 : : {
856 : 140714 : return BufFileSeek(file,
857 : 70357 : (int) (blknum / BUFFILE_SEG_SIZE),
858 : 70357 : (pgoff_t) (blknum % BUFFILE_SEG_SIZE) * BLCKSZ,
859 : : SEEK_SET);
860 : : }
861 : :
862 : : /*
863 : : * Returns the amount of data in the given BufFile, in bytes.
864 : : *
865 : : * Returned value includes the size of any holes left behind by BufFileAppend.
866 : : * ereport()s on failure.
867 : : */
868 : : int64
869 : 291 : BufFileSize(BufFile *file)
870 : : {
871 : : int64 lastFileSize;
872 : :
873 : : /* Get the size of the last physical file. */
874 : 291 : lastFileSize = FileSize(file->files[file->numFiles - 1]);
875 [ - + ]: 291 : if (lastFileSize < 0)
876 [ # # ]: 0 : ereport(ERROR,
877 : : (errcode_for_file_access(),
878 : : errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
879 : : FilePathName(file->files[file->numFiles - 1]),
880 : : file->name)));
881 : :
882 : 291 : return ((file->numFiles - 1) * (int64) MAX_PHYSICAL_FILESIZE) +
883 : : lastFileSize;
884 : : }
885 : :
886 : : /*
887 : : * Append the contents of the source file to the end of the target file.
888 : : *
889 : : * Note that operation subsumes ownership of underlying resources from
890 : : * "source". Caller should never call BufFileClose against source having
891 : : * called here first. Resource owners for source and target must match,
892 : : * too.
893 : : *
894 : : * This operation works by manipulating lists of segment files, so the
895 : : * file content is always appended at a MAX_PHYSICAL_FILESIZE-aligned
896 : : * boundary, typically creating empty holes before the boundary. These
897 : : * areas do not contain any interesting data, and cannot be read from by
898 : : * caller.
899 : : *
900 : : * Returns the block number within target where the contents of source
901 : : * begins. Caller should apply this as an offset when working off block
902 : : * positions that are in terms of the original BufFile space.
903 : : */
904 : : int64
905 : 153 : BufFileAppend(BufFile *target, BufFile *source)
906 : : {
907 : 153 : int64 startBlock = (int64) target->numFiles * BUFFILE_SEG_SIZE;
908 : 153 : int newNumFiles = target->numFiles + source->numFiles;
909 : : int i;
910 : :
911 : : Assert(source->readOnly);
912 : : Assert(!source->dirty);
913 : :
914 [ - + ]: 153 : if (target->resowner != source->resowner)
915 [ # # ]: 0 : elog(ERROR, "could not append BufFile with non-matching resource owner");
916 : :
917 : 153 : target->files = repalloc_array(target->files, File, newNumFiles);
918 [ + + ]: 306 : for (i = target->numFiles; i < newNumFiles; i++)
919 : 153 : target->files[i] = source->files[i - target->numFiles];
920 : 153 : target->numFiles = newNumFiles;
921 : :
922 : 153 : return startBlock;
923 : : }
924 : :
925 : : /*
926 : : * Truncate a BufFile created by BufFileCreateFileSet up to the given fileno
927 : : * and the offset.
928 : : */
929 : : void
930 : 9 : BufFileTruncateFileSet(BufFile *file, int fileno, pgoff_t offset)
931 : : {
932 : 9 : int numFiles = file->numFiles;
933 : 9 : int newFile = fileno;
934 : 9 : pgoff_t newOffset = file->curOffset;
935 : : char segment_name[MAXPGPATH];
936 : : int i;
937 : :
938 : : /*
939 : : * Loop over all the files up to the given fileno and remove the files
940 : : * that are greater than the fileno and truncate the given file up to the
941 : : * offset. Note that we also remove the given fileno if the offset is 0
942 : : * provided it is not the first file in which we truncate it.
943 : : */
944 [ + + ]: 18 : for (i = file->numFiles - 1; i >= fileno; i--)
945 : : {
946 [ + - - + : 9 : if ((i != fileno || offset == 0) && i != 0)
- - ]
947 : : {
948 : 0 : FileSetSegmentName(segment_name, file->name, i);
949 : 0 : FileClose(file->files[i]);
950 [ # # ]: 0 : if (!FileSetDelete(file->fileset, segment_name, true))
951 [ # # ]: 0 : ereport(ERROR,
952 : : (errcode_for_file_access(),
953 : : errmsg("could not delete fileset \"%s\": %m",
954 : : segment_name)));
955 : 0 : numFiles--;
956 : 0 : newOffset = MAX_PHYSICAL_FILESIZE;
957 : :
958 : : /*
959 : : * This is required to indicate that we have deleted the given
960 : : * fileno.
961 : : */
962 [ # # ]: 0 : if (i == fileno)
963 : 0 : newFile--;
964 : : }
965 : : else
966 : : {
967 [ - + ]: 9 : if (FileTruncate(file->files[i], offset,
968 : : WAIT_EVENT_BUFFILE_TRUNCATE) < 0)
969 [ # # ]: 0 : ereport(ERROR,
970 : : (errcode_for_file_access(),
971 : : errmsg("could not truncate file \"%s\": %m",
972 : : FilePathName(file->files[i]))));
973 : 9 : newOffset = offset;
974 : : }
975 : : }
976 : :
977 : 9 : file->numFiles = numFiles;
978 : :
979 : : /*
980 : : * If the truncate point is within existing buffer then we can just adjust
981 : : * pos within buffer.
982 : : */
983 [ + - ]: 9 : if (newFile == file->curFile &&
984 [ + - ]: 9 : newOffset >= file->curOffset &&
985 [ - + ]: 9 : newOffset <= file->curOffset + file->nbytes)
986 : : {
987 : : /* No need to reset the current pos if the new pos is greater. */
988 [ # # ]: 0 : if (newOffset <= file->curOffset + file->pos)
989 : 0 : file->pos = (int64) newOffset - file->curOffset;
990 : :
991 : : /* Adjust the nbytes for the current buffer. */
992 : 0 : file->nbytes = (int64) newOffset - file->curOffset;
993 : : }
994 [ + - ]: 9 : else if (newFile == file->curFile &&
995 [ - + ]: 9 : newOffset < file->curOffset)
996 : : {
997 : : /*
998 : : * The truncate point is within the existing file but prior to the
999 : : * current position, so we can forget the current buffer and reset the
1000 : : * current position.
1001 : : */
1002 : 0 : file->curOffset = newOffset;
1003 : 0 : file->pos = 0;
1004 : 0 : file->nbytes = 0;
1005 : : }
1006 [ - + ]: 9 : else if (newFile < file->curFile)
1007 : : {
1008 : : /*
1009 : : * The truncate point is prior to the current file, so need to reset
1010 : : * the current position accordingly.
1011 : : */
1012 : 0 : file->curFile = newFile;
1013 : 0 : file->curOffset = newOffset;
1014 : 0 : file->pos = 0;
1015 : 0 : file->nbytes = 0;
1016 : : }
1017 : : /* Nothing to do, if the truncate point is beyond current file. */
1018 : 9 : }
|
