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 : 6376 : makeBufFileCommon(int nfiles)
120 : : {
121 : 6376 : BufFile *file = palloc_object(BufFile);
122 : :
123 : 6376 : file->numFiles = nfiles;
124 : 6376 : file->isInterXact = false;
125 : 6376 : file->dirty = false;
126 : 6376 : file->resowner = CurrentResourceOwner;
127 : 6376 : file->curFile = 0;
128 : 6376 : file->curOffset = 0;
129 : 6376 : file->pos = 0;
130 : 6376 : file->nbytes = 0;
131 : :
132 : 6376 : 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 : 2019 : makeBufFile(File firstfile)
141 : : {
142 : 2019 : BufFile *file = makeBufFileCommon(1);
143 : :
144 : 2019 : file->files = palloc_object(File);
145 : 2019 : file->files[0] = firstfile;
146 : 2019 : file->readOnly = false;
147 : 2019 : file->fileset = NULL;
148 : 2019 : file->name = NULL;
149 : :
150 : 2019 : 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 = (File *) repalloc(file->files,
176 : 0 : (file->numFiles + 1) * sizeof(File));
177 : 0 : file->files[file->numFiles] = pfile;
178 : 0 : file->numFiles++;
179 : 0 : }
180 : :
181 : : /*
182 : : * Create a BufFile for a new temporary file (which will expand to become
183 : : * multiple temporary files if more than MAX_PHYSICAL_FILESIZE bytes are
184 : : * written to it).
185 : : *
186 : : * If interXact is true, the temp file will not be automatically deleted
187 : : * at end of transaction.
188 : : *
189 : : * Note: if interXact is true, the caller had better be calling us in a
190 : : * memory context, and with a resource owner, that will survive across
191 : : * transaction boundaries.
192 : : */
193 : : BufFile *
194 : 2019 : BufFileCreateTemp(bool interXact)
195 : : {
196 : : BufFile *file;
197 : : File pfile;
198 : :
199 : : /*
200 : : * Ensure that temp tablespaces are set up for OpenTemporaryFile to use.
201 : : * Possibly the caller will have done this already, but it seems useful to
202 : : * double-check here. Failure to do this at all would result in the temp
203 : : * files always getting placed in the default tablespace, which is a
204 : : * pretty hard-to-detect bug. Callers may prefer to do it earlier if they
205 : : * want to be sure that any required catalog access is done in some other
206 : : * resource context.
207 : : */
208 : 2019 : PrepareTempTablespaces();
209 : :
210 : 2019 : pfile = OpenTemporaryFile(interXact);
211 : : Assert(pfile >= 0);
212 : :
213 : 2019 : file = makeBufFile(pfile);
214 : 2019 : file->isInterXact = interXact;
215 : :
216 : 2019 : return file;
217 : : }
218 : :
219 : : /*
220 : : * Build the name for a given segment of a given BufFile.
221 : : */
222 : : static void
223 : 9378 : FileSetSegmentName(char *name, const char *buffile_name, int segment)
224 : : {
225 : 9378 : snprintf(name, MAXPGPATH, "%s.%d", buffile_name, segment);
226 : 9378 : }
227 : :
228 : : /*
229 : : * Create a new segment file backing a fileset based BufFile.
230 : : */
231 : : static File
232 : 1966 : MakeNewFileSetSegment(BufFile *buffile, int segment)
233 : : {
234 : : char name[MAXPGPATH];
235 : : File file;
236 : :
237 : : /*
238 : : * It is possible that there are files left over from before a crash
239 : : * restart with the same name. In order for BufFileOpenFileSet() not to
240 : : * get confused about how many segments there are, we'll unlink the next
241 : : * segment number if it already exists.
242 : : */
243 : 1966 : FileSetSegmentName(name, buffile->name, segment + 1);
244 : 1966 : FileSetDelete(buffile->fileset, name, true);
245 : :
246 : : /* Create the new segment. */
247 : 1966 : FileSetSegmentName(name, buffile->name, segment);
248 : 1966 : file = FileSetCreate(buffile->fileset, name);
249 : :
250 : : /* FileSetCreate would've errored out */
251 : : Assert(file > 0);
252 : :
253 : 1966 : return file;
254 : : }
255 : :
256 : : /*
257 : : * Create a BufFile that can be discovered and opened read-only by other
258 : : * backends that are attached to the same SharedFileSet using the same name.
259 : : *
260 : : * The naming scheme for fileset based BufFiles is left up to the calling code.
261 : : * The name will appear as part of one or more filenames on disk, and might
262 : : * provide clues to administrators about which subsystem is generating
263 : : * temporary file data. Since each SharedFileSet object is backed by one or
264 : : * more uniquely named temporary directory, names don't conflict with
265 : : * unrelated SharedFileSet objects.
266 : : */
267 : : BufFile *
268 : 1966 : BufFileCreateFileSet(FileSet *fileset, const char *name)
269 : : {
270 : : BufFile *file;
271 : :
272 : 1966 : file = makeBufFileCommon(1);
273 : 1966 : file->fileset = fileset;
274 : 1966 : file->name = pstrdup(name);
275 : 1966 : file->files = palloc_object(File);
276 : 1966 : file->files[0] = MakeNewFileSetSegment(file, 0);
277 : 1966 : file->readOnly = false;
278 : :
279 : 1966 : return file;
280 : : }
281 : :
282 : : /*
283 : : * Open a file that was previously created in another backend (or this one)
284 : : * with BufFileCreateFileSet in the same FileSet using the same name.
285 : : * The backend that created the file must have called BufFileClose() or
286 : : * BufFileExportFileSet() to make sure that it is ready to be opened by other
287 : : * backends and render it read-only. If missing_ok is true, which indicates
288 : : * that missing files can be safely ignored, then return NULL if the BufFile
289 : : * with the given name is not found, otherwise, throw an error.
290 : : */
291 : : BufFile *
292 : 2664 : BufFileOpenFileSet(FileSet *fileset, const char *name, int mode,
293 : : bool missing_ok)
294 : : {
295 : : BufFile *file;
296 : : char segment_name[MAXPGPATH];
297 : 2664 : Size capacity = 16;
298 : : File *files;
299 : 2664 : int nfiles = 0;
300 : :
301 : 2664 : files = palloc_array(File, capacity);
302 : :
303 : : /*
304 : : * We don't know how many segments there are, so we'll probe the
305 : : * filesystem to find out.
306 : : */
307 : : for (;;)
308 : : {
309 : : /* See if we need to expand our file segment array. */
310 [ - + ]: 5055 : if (nfiles + 1 > capacity)
311 : : {
312 : 0 : capacity *= 2;
313 : 0 : files = repalloc_array(files, File, capacity);
314 : : }
315 : : /* Try to load a segment. */
316 : 5055 : FileSetSegmentName(segment_name, name, nfiles);
317 : 5055 : files[nfiles] = FileSetOpen(fileset, segment_name, mode);
318 [ + + ]: 5055 : if (files[nfiles] <= 0)
319 : 2664 : break;
320 : 2391 : ++nfiles;
321 : :
322 [ + + ]: 2391 : CHECK_FOR_INTERRUPTS();
323 : : }
324 : :
325 : : /*
326 : : * If we didn't find any files at all, then no BufFile exists with this
327 : : * name.
328 : : */
329 [ + + ]: 2664 : if (nfiles == 0)
330 : : {
331 : : /* free the memory */
332 : 273 : pfree(files);
333 : :
334 [ + - ]: 273 : if (missing_ok)
335 : 273 : return NULL;
336 : :
337 [ # # ]: 0 : ereport(ERROR,
338 : : (errcode_for_file_access(),
339 : : errmsg("could not open temporary file \"%s\" from BufFile \"%s\": %m",
340 : : segment_name, name)));
341 : : }
342 : :
343 : 2391 : file = makeBufFileCommon(nfiles);
344 : 2391 : file->files = files;
345 : 2391 : file->readOnly = (mode == O_RDONLY);
346 : 2391 : file->fileset = fileset;
347 : 2391 : file->name = pstrdup(name);
348 : :
349 : 2391 : return file;
350 : : }
351 : :
352 : : /*
353 : : * Delete a BufFile that was created by BufFileCreateFileSet in the given
354 : : * FileSet using the given name.
355 : : *
356 : : * It is not necessary to delete files explicitly with this function. It is
357 : : * provided only as a way to delete files proactively, rather than waiting for
358 : : * the FileSet to be cleaned up.
359 : : *
360 : : * Only one backend should attempt to delete a given name, and should know
361 : : * that it exists and has been exported or closed otherwise missing_ok should
362 : : * be passed true.
363 : : */
364 : : void
365 : 352 : BufFileDeleteFileSet(FileSet *fileset, const char *name, bool missing_ok)
366 : : {
367 : : char segment_name[MAXPGPATH];
368 : 352 : int segment = 0;
369 : 352 : bool found = false;
370 : :
371 : : /*
372 : : * We don't know how many segments the file has. We'll keep deleting
373 : : * until we run out. If we don't manage to find even an initial segment,
374 : : * raise an error.
375 : : */
376 : : for (;;)
377 : : {
378 : 391 : FileSetSegmentName(segment_name, name, segment);
379 [ + + ]: 391 : if (!FileSetDelete(fileset, segment_name, true))
380 : 352 : break;
381 : 39 : found = true;
382 : 39 : ++segment;
383 : :
384 [ - + ]: 39 : CHECK_FOR_INTERRUPTS();
385 : : }
386 : :
387 [ + + - + ]: 352 : if (!found && !missing_ok)
388 [ # # ]: 0 : elog(ERROR, "could not delete unknown BufFile \"%s\"", name);
389 : 352 : }
390 : :
391 : : /*
392 : : * BufFileExportFileSet --- flush and make read-only, in preparation for sharing.
393 : : */
394 : : void
395 : 379 : BufFileExportFileSet(BufFile *file)
396 : : {
397 : : /* Must be a file belonging to a FileSet. */
398 : : Assert(file->fileset != NULL);
399 : :
400 : : /* It's probably a bug if someone calls this twice. */
401 : : Assert(!file->readOnly);
402 : :
403 : 379 : BufFileFlush(file);
404 : 379 : file->readOnly = true;
405 : 379 : }
406 : :
407 : : /*
408 : : * Close a BufFile
409 : : *
410 : : * Like fclose(), this also implicitly FileCloses the underlying File.
411 : : */
412 : : void
413 : 6216 : BufFileClose(BufFile *file)
414 : : {
415 : : int i;
416 : :
417 : : /* flush any unwritten data */
418 : 6216 : BufFileFlush(file);
419 : : /* close and delete the underlying file(s) */
420 [ + + ]: 12585 : for (i = 0; i < file->numFiles; i++)
421 : 6369 : FileClose(file->files[i]);
422 : : /* release the buffer space */
423 : 6216 : pfree(file->files);
424 : 6216 : pfree(file);
425 : 6216 : }
426 : :
427 : : /*
428 : : * BufFileLoadBuffer
429 : : *
430 : : * Load some data into buffer, if possible, starting from curOffset.
431 : : * At call, must have dirty = false, pos and nbytes = 0.
432 : : * On exit, nbytes is number of bytes loaded.
433 : : */
434 : : static void
435 : 64068 : BufFileLoadBuffer(BufFile *file)
436 : : {
437 : : File thisfile;
438 : : instr_time io_start;
439 : : instr_time io_time;
440 : : ssize_t rc;
441 : :
442 : : /*
443 : : * Advance to next component file if necessary and possible.
444 : : */
445 [ - + ]: 64068 : if (file->curOffset >= MAX_PHYSICAL_FILESIZE &&
446 [ # # ]: 0 : file->curFile + 1 < file->numFiles)
447 : : {
448 : 0 : file->curFile++;
449 : 0 : file->curOffset = 0;
450 : : }
451 : :
452 : 64068 : thisfile = file->files[file->curFile];
453 : :
454 [ - + ]: 64068 : if (track_io_timing)
455 : 0 : INSTR_TIME_SET_CURRENT(io_start);
456 : : else
457 : 64068 : INSTR_TIME_SET_ZERO(io_start);
458 : :
459 : : /*
460 : : * Read whatever we can get, up to a full bufferload.
461 : : */
462 : 64068 : rc = FileRead(thisfile,
463 : 64068 : file->buffer.data,
464 : : sizeof(file->buffer.data),
465 : : file->curOffset,
466 : : WAIT_EVENT_BUFFILE_READ);
467 [ - + ]: 64068 : if (rc < 0)
468 : : {
469 : 0 : file->nbytes = 0;
470 [ # # ]: 0 : ereport(ERROR,
471 : : (errcode_for_file_access(),
472 : : errmsg("could not read file \"%s\": %m",
473 : : FilePathName(thisfile))));
474 : : }
475 : :
476 : 64068 : file->nbytes = rc;
477 : :
478 [ - + ]: 64068 : if (track_io_timing)
479 : : {
480 : 0 : INSTR_TIME_SET_CURRENT(io_time);
481 : 0 : INSTR_TIME_ACCUM_DIFF(pgBufferUsage.temp_blk_read_time, io_time, io_start);
482 : : }
483 : :
484 : : /* we choose not to advance curOffset here */
485 : :
486 [ + + ]: 64068 : if (file->nbytes > 0)
487 : 62322 : pgBufferUsage.temp_blks_read++;
488 : 64068 : }
489 : :
490 : : /*
491 : : * BufFileDumpBuffer
492 : : *
493 : : * Dump buffer contents starting at curOffset.
494 : : * At call, should have dirty = true, nbytes > 0.
495 : : * On exit, dirty is cleared if successful write, and curOffset is advanced.
496 : : */
497 : : static void
498 : 69307 : BufFileDumpBuffer(BufFile *file)
499 : : {
500 : 69307 : int64 wpos = 0;
501 : : File thisfile;
502 : :
503 : : /*
504 : : * Unlike BufFileLoadBuffer, we must dump the whole buffer even if it
505 : : * crosses a component-file boundary; so we need a loop.
506 : : */
507 [ + + ]: 138614 : while (wpos < file->nbytes)
508 : : {
509 : : int64 availbytes;
510 : : instr_time io_start;
511 : : instr_time io_time;
512 : : size_t bytestowrite;
513 : : ssize_t rc;
514 : :
515 : : /*
516 : : * Advance to next component file if necessary and possible.
517 : : */
518 [ - + ]: 69307 : if (file->curOffset >= MAX_PHYSICAL_FILESIZE)
519 : : {
520 [ # # ]: 0 : while (file->curFile + 1 >= file->numFiles)
521 : 0 : extendBufFile(file);
522 : 0 : file->curFile++;
523 : 0 : file->curOffset = 0;
524 : : }
525 : :
526 : : /*
527 : : * Determine how much we need to write into this file.
528 : : */
529 : 69307 : bytestowrite = file->nbytes - wpos;
530 : 69307 : availbytes = MAX_PHYSICAL_FILESIZE - file->curOffset;
531 : :
532 [ - + ]: 69307 : if (bytestowrite > availbytes)
533 : 0 : bytestowrite = availbytes;
534 : :
535 : 69307 : thisfile = file->files[file->curFile];
536 : :
537 [ - + ]: 69307 : if (track_io_timing)
538 : 0 : INSTR_TIME_SET_CURRENT(io_start);
539 : : else
540 : 69307 : INSTR_TIME_SET_ZERO(io_start);
541 : :
542 : 69307 : rc = FileWrite(thisfile,
543 : 69307 : file->buffer.data + wpos,
544 : : bytestowrite,
545 : : file->curOffset,
546 : : WAIT_EVENT_BUFFILE_WRITE);
547 [ - + ]: 69307 : if (rc <= 0)
548 [ # # ]: 0 : ereport(ERROR,
549 : : (errcode_for_file_access(),
550 : : errmsg("could not write to file \"%s\": %m",
551 : : FilePathName(thisfile))));
552 : :
553 [ - + ]: 69307 : if (track_io_timing)
554 : : {
555 : 0 : INSTR_TIME_SET_CURRENT(io_time);
556 : 0 : INSTR_TIME_ACCUM_DIFF(pgBufferUsage.temp_blk_write_time, io_time, io_start);
557 : : }
558 : :
559 : 69307 : file->curOffset += rc;
560 : 69307 : wpos += rc;
561 : :
562 : 69307 : pgBufferUsage.temp_blks_written++;
563 : : }
564 : 69307 : file->dirty = false;
565 : :
566 : : /*
567 : : * At this point, curOffset has been advanced to the end of the buffer,
568 : : * ie, its original value + nbytes. We need to make it point to the
569 : : * logical file position, ie, original value + pos, in case that is less
570 : : * (as could happen due to a small backwards seek in a dirty buffer!)
571 : : */
572 : 69307 : file->curOffset -= (file->nbytes - file->pos);
573 [ - + ]: 69307 : if (file->curOffset < 0) /* handle possible segment crossing */
574 : : {
575 : 0 : file->curFile--;
576 : : Assert(file->curFile >= 0);
577 : 0 : file->curOffset += MAX_PHYSICAL_FILESIZE;
578 : : }
579 : :
580 : : /*
581 : : * Now we can set the buffer empty without changing the logical position
582 : : */
583 : 69307 : file->pos = 0;
584 : 69307 : file->nbytes = 0;
585 : 69307 : }
586 : :
587 : : /*
588 : : * BufFileRead variants
589 : : *
590 : : * Like fread() except we assume 1-byte element size and report I/O errors via
591 : : * ereport().
592 : : *
593 : : * If 'exact' is true, then an error is also raised if the number of bytes
594 : : * read is not exactly 'size' (no short reads). If 'exact' and 'eofOK' are
595 : : * true, then reading zero bytes is ok.
596 : : */
597 : : static size_t
598 : 16156450 : BufFileReadCommon(BufFile *file, void *ptr, size_t size, bool exact, bool eofOK)
599 : : {
600 : 16156450 : size_t start_size = size;
601 : 16156450 : size_t nread = 0;
602 : : size_t nthistime;
603 : :
604 : 16156450 : BufFileFlush(file);
605 : :
606 [ + + ]: 32327528 : while (size > 0)
607 : : {
608 [ + + ]: 16172824 : if (file->pos >= file->nbytes)
609 : : {
610 : : /* Try to load more data into buffer. */
611 : 64068 : file->curOffset += file->pos;
612 : 64068 : file->pos = 0;
613 : 64068 : file->nbytes = 0;
614 : 64068 : BufFileLoadBuffer(file);
615 [ + + ]: 64068 : if (file->nbytes <= 0)
616 : 1746 : break; /* no more data available */
617 : : }
618 : :
619 : 16171078 : nthistime = file->nbytes - file->pos;
620 [ + + ]: 16171078 : if (nthistime > size)
621 : 16110769 : nthistime = size;
622 : : Assert(nthistime > 0);
623 : :
624 : 16171078 : memcpy(ptr, file->buffer.data + file->pos, nthistime);
625 : :
626 : 16171078 : file->pos += nthistime;
627 : 16171078 : ptr = (char *) ptr + nthistime;
628 : 16171078 : size -= nthistime;
629 : 16171078 : nread += nthistime;
630 : : }
631 : :
632 [ + - + + ]: 16156450 : if (exact &&
633 [ + - - + ]: 1746 : (nread != start_size && !(nread == 0 && eofOK)))
634 [ # # # # ]: 0 : ereport(ERROR,
635 : : errcode_for_file_access(),
636 : : file->name ?
637 : : errmsg("could not read from file set \"%s\": read only %zu of %zu bytes",
638 : : file->name, nread, start_size) :
639 : : errmsg("could not read from temporary file: read only %zu of %zu bytes",
640 : : nread, start_size));
641 : :
642 : 16156450 : return nread;
643 : : }
644 : :
645 : : /*
646 : : * Legacy interface where the caller needs to check for end of file or short
647 : : * reads.
648 : : */
649 : : size_t
650 : 0 : BufFileRead(BufFile *file, void *ptr, size_t size)
651 : : {
652 : 0 : return BufFileReadCommon(file, ptr, size, false, false);
653 : : }
654 : :
655 : : /*
656 : : * Require read of exactly the specified size.
657 : : */
658 : : void
659 : 10539595 : BufFileReadExact(BufFile *file, void *ptr, size_t size)
660 : : {
661 : 10539595 : BufFileReadCommon(file, ptr, size, true, false);
662 : 10539595 : }
663 : :
664 : : /*
665 : : * Require read of exactly the specified size, but optionally allow end of
666 : : * file (in which case 0 is returned).
667 : : */
668 : : size_t
669 : 5616855 : BufFileReadMaybeEOF(BufFile *file, void *ptr, size_t size, bool eofOK)
670 : : {
671 : 5616855 : return BufFileReadCommon(file, ptr, size, true, eofOK);
672 : : }
673 : :
674 : : /*
675 : : * BufFileWrite
676 : : *
677 : : * Like fwrite() except we assume 1-byte element size and report errors via
678 : : * ereport().
679 : : */
680 : : void
681 : 11669169 : BufFileWrite(BufFile *file, const void *ptr, size_t size)
682 : : {
683 : : size_t nthistime;
684 : :
685 : : Assert(!file->readOnly);
686 : :
687 [ + + ]: 23363260 : while (size > 0)
688 : : {
689 [ + + ]: 11694091 : if (file->pos >= BLCKSZ)
690 : : {
691 : : /* Buffer full, dump it out */
692 [ + + ]: 42306 : if (file->dirty)
693 : 41990 : BufFileDumpBuffer(file);
694 : : else
695 : : {
696 : : /* Hmm, went directly from reading to writing? */
697 : 316 : file->curOffset += file->pos;
698 : 316 : file->pos = 0;
699 : 316 : file->nbytes = 0;
700 : : }
701 : : }
702 : :
703 : 11694091 : nthistime = BLCKSZ - file->pos;
704 [ + + ]: 11694091 : if (nthistime > size)
705 : 11627115 : nthistime = size;
706 : : Assert(nthistime > 0);
707 : :
708 : 11694091 : memcpy(file->buffer.data + file->pos, ptr, nthistime);
709 : :
710 : 11694091 : file->dirty = true;
711 : 11694091 : file->pos += nthistime;
712 [ + + ]: 11694091 : if (file->nbytes < file->pos)
713 : 11691515 : file->nbytes = file->pos;
714 : 11694091 : ptr = (const char *) ptr + nthistime;
715 : 11694091 : size -= nthistime;
716 : : }
717 : 11669169 : }
718 : :
719 : : /*
720 : : * BufFileFlush
721 : : *
722 : : * Like fflush(), except that I/O errors are reported with ereport().
723 : : */
724 : : static void
725 : 16198690 : BufFileFlush(BufFile *file)
726 : : {
727 [ + + ]: 16198690 : if (file->dirty)
728 : 27317 : BufFileDumpBuffer(file);
729 : :
730 : : Assert(!file->dirty);
731 : 16198690 : }
732 : :
733 : : /*
734 : : * BufFileSeek
735 : : *
736 : : * Like fseek(), except that target position needs two values in order to
737 : : * work when logical filesize exceeds maximum value representable by pgoff_t.
738 : : * We do not support relative seeks across more than that, however.
739 : : * I/O errors are reported by ereport().
740 : : *
741 : : * Result is 0 if OK, EOF if not. Logical position is not moved if an
742 : : * impossible seek is attempted.
743 : : */
744 : : int
745 : 72586 : BufFileSeek(BufFile *file, int fileno, pgoff_t offset, int whence)
746 : : {
747 : : int newFile;
748 : : pgoff_t newOffset;
749 : :
750 [ + - + - ]: 72586 : switch (whence)
751 : : {
752 : 72251 : case SEEK_SET:
753 [ - + ]: 72251 : if (fileno < 0)
754 : 0 : return EOF;
755 : 72251 : newFile = fileno;
756 : 72251 : newOffset = offset;
757 : 72251 : break;
758 : 0 : case SEEK_CUR:
759 : :
760 : : /*
761 : : * Relative seek considers only the signed offset, ignoring
762 : : * fileno.
763 : : */
764 : 0 : newFile = file->curFile;
765 : 0 : newOffset = (file->curOffset + file->pos) + offset;
766 : 0 : break;
767 : 335 : case SEEK_END:
768 : :
769 : : /*
770 : : * The file size of the last file gives us the end offset of that
771 : : * file.
772 : : */
773 : 335 : newFile = file->numFiles - 1;
774 : 335 : newOffset = FileSize(file->files[file->numFiles - 1]);
775 [ - + ]: 335 : if (newOffset < 0)
776 [ # # ]: 0 : ereport(ERROR,
777 : : (errcode_for_file_access(),
778 : : errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
779 : : FilePathName(file->files[file->numFiles - 1]),
780 : : file->name)));
781 : 335 : break;
782 : 0 : default:
783 [ # # ]: 0 : elog(ERROR, "invalid whence: %d", whence);
784 : : return EOF;
785 : : }
786 [ - + ]: 72586 : while (newOffset < 0)
787 : : {
788 [ # # ]: 0 : if (--newFile < 0)
789 : 0 : return EOF;
790 : 0 : newOffset += MAX_PHYSICAL_FILESIZE;
791 : : }
792 [ + + ]: 72586 : if (newFile == file->curFile &&
793 [ + + ]: 72433 : newOffset >= file->curOffset &&
794 [ + + ]: 52965 : newOffset <= file->curOffset + file->nbytes)
795 : : {
796 : : /*
797 : : * Seek is to a point within existing buffer; we can just adjust
798 : : * pos-within-buffer, without flushing buffer. Note this is OK
799 : : * whether reading or writing, but buffer remains dirty if we were
800 : : * writing.
801 : : */
802 : 36941 : file->pos = (int64) (newOffset - file->curOffset);
803 : 36941 : return 0;
804 : : }
805 : : /* Otherwise, must reposition buffer, so flush any dirty data */
806 : 35645 : BufFileFlush(file);
807 : :
808 : : /*
809 : : * At this point and no sooner, check for seek past last segment. The
810 : : * above flush could have created a new segment, so checking sooner would
811 : : * not work (at least not with this code).
812 : : */
813 : :
814 : : /* convert seek to "start of next seg" to "end of last seg" */
815 [ - + - - ]: 35645 : if (newFile == file->numFiles && newOffset == 0)
816 : : {
817 : 0 : newFile--;
818 : 0 : newOffset = MAX_PHYSICAL_FILESIZE;
819 : : }
820 [ - + ]: 35645 : while (newOffset > MAX_PHYSICAL_FILESIZE)
821 : : {
822 [ # # ]: 0 : if (++newFile >= file->numFiles)
823 : 0 : return EOF;
824 : 0 : newOffset -= MAX_PHYSICAL_FILESIZE;
825 : : }
826 [ - + ]: 35645 : if (newFile >= file->numFiles)
827 : 0 : return EOF;
828 : : /* Seek is OK! */
829 : 35645 : file->curFile = newFile;
830 : 35645 : file->curOffset = newOffset;
831 : 35645 : file->pos = 0;
832 : 35645 : file->nbytes = 0;
833 : 35645 : return 0;
834 : : }
835 : :
836 : : void
837 : 88672 : BufFileTell(BufFile *file, int *fileno, pgoff_t *offset)
838 : : {
839 : 88672 : *fileno = file->curFile;
840 : 88672 : *offset = file->curOffset + file->pos;
841 : 88672 : }
842 : :
843 : : /*
844 : : * BufFileSeekBlock --- block-oriented seek
845 : : *
846 : : * Performs absolute seek to the start of the n'th BLCKSZ-sized block of
847 : : * the file. Note that users of this interface will fail if their files
848 : : * exceed BLCKSZ * PG_INT64_MAX bytes, but that is quite a lot; we don't
849 : : * work with tables bigger than that, either...
850 : : *
851 : : * Result is 0 if OK, EOF if not. Logical position is not moved if an
852 : : * impossible seek is attempted.
853 : : */
854 : : int
855 : 70351 : BufFileSeekBlock(BufFile *file, int64 blknum)
856 : : {
857 : 140702 : return BufFileSeek(file,
858 : 70351 : (int) (blknum / BUFFILE_SEG_SIZE),
859 : 70351 : (pgoff_t) (blknum % BUFFILE_SEG_SIZE) * BLCKSZ,
860 : : SEEK_SET);
861 : : }
862 : :
863 : : /*
864 : : * Returns the amount of data in the given BufFile, in bytes.
865 : : *
866 : : * Returned value includes the size of any holes left behind by BufFileAppend.
867 : : * ereport()s on failure.
868 : : */
869 : : int64
870 : 291 : BufFileSize(BufFile *file)
871 : : {
872 : : int64 lastFileSize;
873 : :
874 : : /* Get the size of the last physical file. */
875 : 291 : lastFileSize = FileSize(file->files[file->numFiles - 1]);
876 [ - + ]: 291 : if (lastFileSize < 0)
877 [ # # ]: 0 : ereport(ERROR,
878 : : (errcode_for_file_access(),
879 : : errmsg("could not determine size of temporary file \"%s\" from BufFile \"%s\": %m",
880 : : FilePathName(file->files[file->numFiles - 1]),
881 : : file->name)));
882 : :
883 : 291 : return ((file->numFiles - 1) * (int64) MAX_PHYSICAL_FILESIZE) +
884 : : lastFileSize;
885 : : }
886 : :
887 : : /*
888 : : * Append the contents of the source file to the end of the target file.
889 : : *
890 : : * Note that operation subsumes ownership of underlying resources from
891 : : * "source". Caller should never call BufFileClose against source having
892 : : * called here first. Resource owners for source and target must match,
893 : : * too.
894 : : *
895 : : * This operation works by manipulating lists of segment files, so the
896 : : * file content is always appended at a MAX_PHYSICAL_FILESIZE-aligned
897 : : * boundary, typically creating empty holes before the boundary. These
898 : : * areas do not contain any interesting data, and cannot be read from by
899 : : * caller.
900 : : *
901 : : * Returns the block number within target where the contents of source
902 : : * begins. Caller should apply this as an offset when working off block
903 : : * positions that are in terms of the original BufFile space.
904 : : */
905 : : int64
906 : 153 : BufFileAppend(BufFile *target, BufFile *source)
907 : : {
908 : 153 : int64 startBlock = (int64) target->numFiles * BUFFILE_SEG_SIZE;
909 : 153 : int newNumFiles = target->numFiles + source->numFiles;
910 : : int i;
911 : :
912 : : Assert(source->readOnly);
913 : : Assert(!source->dirty);
914 : :
915 [ - + ]: 153 : if (target->resowner != source->resowner)
916 [ # # ]: 0 : elog(ERROR, "could not append BufFile with non-matching resource owner");
917 : :
918 : 153 : target->files = (File *)
919 : 153 : repalloc(target->files, sizeof(File) * newNumFiles);
920 [ + + ]: 306 : for (i = target->numFiles; i < newNumFiles; i++)
921 : 153 : target->files[i] = source->files[i - target->numFiles];
922 : 153 : target->numFiles = newNumFiles;
923 : :
924 : 153 : return startBlock;
925 : : }
926 : :
927 : : /*
928 : : * Truncate a BufFile created by BufFileCreateFileSet up to the given fileno
929 : : * and the offset.
930 : : */
931 : : void
932 : 9 : BufFileTruncateFileSet(BufFile *file, int fileno, pgoff_t offset)
933 : : {
934 : 9 : int numFiles = file->numFiles;
935 : 9 : int newFile = fileno;
936 : 9 : pgoff_t newOffset = file->curOffset;
937 : : char segment_name[MAXPGPATH];
938 : : int i;
939 : :
940 : : /*
941 : : * Loop over all the files up to the given fileno and remove the files
942 : : * that are greater than the fileno and truncate the given file up to the
943 : : * offset. Note that we also remove the given fileno if the offset is 0
944 : : * provided it is not the first file in which we truncate it.
945 : : */
946 [ + + ]: 18 : for (i = file->numFiles - 1; i >= fileno; i--)
947 : : {
948 [ + - - + : 9 : if ((i != fileno || offset == 0) && i != 0)
- - ]
949 : : {
950 : 0 : FileSetSegmentName(segment_name, file->name, i);
951 : 0 : FileClose(file->files[i]);
952 [ # # ]: 0 : if (!FileSetDelete(file->fileset, segment_name, true))
953 [ # # ]: 0 : ereport(ERROR,
954 : : (errcode_for_file_access(),
955 : : errmsg("could not delete fileset \"%s\": %m",
956 : : segment_name)));
957 : 0 : numFiles--;
958 : 0 : newOffset = MAX_PHYSICAL_FILESIZE;
959 : :
960 : : /*
961 : : * This is required to indicate that we have deleted the given
962 : : * fileno.
963 : : */
964 [ # # ]: 0 : if (i == fileno)
965 : 0 : newFile--;
966 : : }
967 : : else
968 : : {
969 [ - + ]: 9 : if (FileTruncate(file->files[i], offset,
970 : : WAIT_EVENT_BUFFILE_TRUNCATE) < 0)
971 [ # # ]: 0 : ereport(ERROR,
972 : : (errcode_for_file_access(),
973 : : errmsg("could not truncate file \"%s\": %m",
974 : : FilePathName(file->files[i]))));
975 : 9 : newOffset = offset;
976 : : }
977 : : }
978 : :
979 : 9 : file->numFiles = numFiles;
980 : :
981 : : /*
982 : : * If the truncate point is within existing buffer then we can just adjust
983 : : * pos within buffer.
984 : : */
985 [ + - ]: 9 : if (newFile == file->curFile &&
986 [ + - ]: 9 : newOffset >= file->curOffset &&
987 [ - + ]: 9 : newOffset <= file->curOffset + file->nbytes)
988 : : {
989 : : /* No need to reset the current pos if the new pos is greater. */
990 [ # # ]: 0 : if (newOffset <= file->curOffset + file->pos)
991 : 0 : file->pos = (int64) newOffset - file->curOffset;
992 : :
993 : : /* Adjust the nbytes for the current buffer. */
994 : 0 : file->nbytes = (int64) newOffset - file->curOffset;
995 : : }
996 [ + - ]: 9 : else if (newFile == file->curFile &&
997 [ - + ]: 9 : newOffset < file->curOffset)
998 : : {
999 : : /*
1000 : : * The truncate point is within the existing file but prior to the
1001 : : * current position, so we can forget the current buffer and reset the
1002 : : * current position.
1003 : : */
1004 : 0 : file->curOffset = newOffset;
1005 : 0 : file->pos = 0;
1006 : 0 : file->nbytes = 0;
1007 : : }
1008 [ - + ]: 9 : else if (newFile < file->curFile)
1009 : : {
1010 : : /*
1011 : : * The truncate point is prior to the current file, so need to reset
1012 : : * the current position accordingly.
1013 : : */
1014 : 0 : file->curFile = newFile;
1015 : 0 : file->curOffset = newOffset;
1016 : 0 : file->pos = 0;
1017 : 0 : file->nbytes = 0;
1018 : : }
1019 : : /* Nothing to do, if the truncate point is beyond current file. */
1020 : 9 : }
|
