PostgreSQL Source Code  git master
md.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * md.c
4  * This code manages relations that reside on magnetic disk.
5  *
6  * Or at least, that was what the Berkeley folk had in mind when they named
7  * this file. In reality, what this code provides is an interface from
8  * the smgr API to Unix-like filesystem APIs, so it will work with any type
9  * of device for which the operating system provides filesystem support.
10  * It doesn't matter whether the bits are on spinning rust or some other
11  * storage technology.
12  *
13  * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
14  * Portions Copyright (c) 1994, Regents of the University of California
15  *
16  *
17  * IDENTIFICATION
18  * src/backend/storage/smgr/md.c
19  *
20  *-------------------------------------------------------------------------
21  */
22 #include "postgres.h"
23 
24 #include <unistd.h>
25 #include <fcntl.h>
26 #include <sys/file.h>
27 
28 #include "access/xlog.h"
29 #include "access/xlogutils.h"
30 #include "commands/tablespace.h"
31 #include "miscadmin.h"
32 #include "pg_trace.h"
33 #include "pgstat.h"
34 #include "postmaster/bgwriter.h"
35 #include "storage/bufmgr.h"
36 #include "storage/fd.h"
37 #include "storage/md.h"
38 #include "storage/relfilenode.h"
39 #include "storage/smgr.h"
40 #include "storage/sync.h"
41 #include "utils/hsearch.h"
42 #include "utils/memutils.h"
43 
44 /*
45  * The magnetic disk storage manager keeps track of open file
46  * descriptors in its own descriptor pool. This is done to make it
47  * easier to support relations that are larger than the operating
48  * system's file size limit (often 2GBytes). In order to do that,
49  * we break relations up into "segment" files that are each shorter than
50  * the OS file size limit. The segment size is set by the RELSEG_SIZE
51  * configuration constant in pg_config.h.
52  *
53  * On disk, a relation must consist of consecutively numbered segment
54  * files in the pattern
55  * -- Zero or more full segments of exactly RELSEG_SIZE blocks each
56  * -- Exactly one partial segment of size 0 <= size < RELSEG_SIZE blocks
57  * -- Optionally, any number of inactive segments of size 0 blocks.
58  * The full and partial segments are collectively the "active" segments.
59  * Inactive segments are those that once contained data but are currently
60  * not needed because of an mdtruncate() operation. The reason for leaving
61  * them present at size zero, rather than unlinking them, is that other
62  * backends and/or the checkpointer might be holding open file references to
63  * such segments. If the relation expands again after mdtruncate(), such
64  * that a deactivated segment becomes active again, it is important that
65  * such file references still be valid --- else data might get written
66  * out to an unlinked old copy of a segment file that will eventually
67  * disappear.
68  *
69  * File descriptors are stored in the per-fork md_seg_fds arrays inside
70  * SMgrRelation. The length of these arrays is stored in md_num_open_segs.
71  * Note that a fork's md_num_open_segs having a specific value does not
72  * necessarily mean the relation doesn't have additional segments; we may
73  * just not have opened the next segment yet. (We could not have "all
74  * segments are in the array" as an invariant anyway, since another backend
75  * could extend the relation while we aren't looking.) We do not have
76  * entries for inactive segments, however; as soon as we find a partial
77  * segment, we assume that any subsequent segments are inactive.
78  *
79  * The entire MdfdVec array is palloc'd in the MdCxt memory context.
80  */
81 
82 typedef struct _MdfdVec
83 {
84  File mdfd_vfd; /* fd number in fd.c's pool */
85  BlockNumber mdfd_segno; /* segment number, from 0 */
86 } MdfdVec;
87 
88 static MemoryContext MdCxt; /* context for all MdfdVec objects */
89 
90 
91 /* Populate a file tag describing an md.c segment file. */
92 #define INIT_MD_FILETAG(a,xx_rnode,xx_forknum,xx_segno) \
93 ( \
94  memset(&(a), 0, sizeof(FileTag)), \
95  (a).handler = SYNC_HANDLER_MD, \
96  (a).rnode = (xx_rnode), \
97  (a).forknum = (xx_forknum), \
98  (a).segno = (xx_segno) \
99 )
100 
101 
102 /*** behavior for mdopen & _mdfd_getseg ***/
103 /* ereport if segment not present */
104 #define EXTENSION_FAIL (1 << 0)
105 /* return NULL if segment not present */
106 #define EXTENSION_RETURN_NULL (1 << 1)
107 /* create new segments as needed */
108 #define EXTENSION_CREATE (1 << 2)
109 /* create new segments if needed during recovery */
110 #define EXTENSION_CREATE_RECOVERY (1 << 3)
111 /*
112  * Allow opening segments which are preceded by segments smaller than
113  * RELSEG_SIZE, e.g. inactive segments (see above). Note that this breaks
114  * mdnblocks() and related functionality henceforth - which currently is ok,
115  * because this is only required in the checkpointer which never uses
116  * mdnblocks().
117  */
118 #define EXTENSION_DONT_CHECK_SIZE (1 << 4)
119 
120 
121 /* local routines */
122 static void mdunlinkfork(RelFileNodeBackend rnode, ForkNumber forkNum,
123  bool isRedo);
124 static MdfdVec *mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior);
125 static void register_dirty_segment(SMgrRelation reln, ForkNumber forknum,
126  MdfdVec *seg);
127 static void register_unlink_segment(RelFileNodeBackend rnode, ForkNumber forknum,
128  BlockNumber segno);
129 static void register_forget_request(RelFileNodeBackend rnode, ForkNumber forknum,
130  BlockNumber segno);
131 static void _fdvec_resize(SMgrRelation reln,
132  ForkNumber forknum,
133  int nseg);
134 static char *_mdfd_segpath(SMgrRelation reln, ForkNumber forknum,
135  BlockNumber segno);
136 static MdfdVec *_mdfd_openseg(SMgrRelation reln, ForkNumber forkno,
137  BlockNumber segno, int oflags);
138 static MdfdVec *_mdfd_getseg(SMgrRelation reln, ForkNumber forkno,
139  BlockNumber blkno, bool skipFsync, int behavior);
140 static BlockNumber _mdnblocks(SMgrRelation reln, ForkNumber forknum,
141  MdfdVec *seg);
142 
143 
144 /*
145  * mdinit() -- Initialize private state for magnetic disk storage manager.
146  */
147 void
148 mdinit(void)
149 {
151  "MdSmgr",
153 }
154 
155 /*
156  * mdexists() -- Does the physical file exist?
157  *
158  * Note: this will return true for lingering files, with pending deletions
159  */
160 bool
162 {
163  /*
164  * Close it first, to ensure that we notice if the fork has been unlinked
165  * since we opened it.
166  */
167  mdclose(reln, forkNum);
168 
169  return (mdopenfork(reln, forkNum, EXTENSION_RETURN_NULL) != NULL);
170 }
171 
172 /*
173  * mdcreate() -- Create a new relation on magnetic disk.
174  *
175  * If isRedo is true, it's okay for the relation to exist already.
176  */
177 void
178 mdcreate(SMgrRelation reln, ForkNumber forkNum, bool isRedo)
179 {
180  MdfdVec *mdfd;
181  char *path;
182  File fd;
183 
184  if (isRedo && reln->md_num_open_segs[forkNum] > 0)
185  return; /* created and opened already... */
186 
187  Assert(reln->md_num_open_segs[forkNum] == 0);
188 
189  /*
190  * We may be using the target table space for the first time in this
191  * database, so create a per-database subdirectory if needed.
192  *
193  * XXX this is a fairly ugly violation of module layering, but this seems
194  * to be the best place to put the check. Maybe TablespaceCreateDbspace
195  * should be here and not in commands/tablespace.c? But that would imply
196  * importing a lot of stuff that smgr.c oughtn't know, either.
197  */
199  reln->smgr_rnode.node.dbNode,
200  isRedo);
201 
202  path = relpath(reln->smgr_rnode, forkNum);
203 
204  fd = PathNameOpenFile(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
205 
206  if (fd < 0)
207  {
208  int save_errno = errno;
209 
210  if (isRedo)
211  fd = PathNameOpenFile(path, O_RDWR | PG_BINARY);
212  if (fd < 0)
213  {
214  /* be sure to report the error reported by create, not open */
215  errno = save_errno;
216  ereport(ERROR,
218  errmsg("could not create file \"%s\": %m", path)));
219  }
220  }
221 
222  pfree(path);
223 
224  _fdvec_resize(reln, forkNum, 1);
225  mdfd = &reln->md_seg_fds[forkNum][0];
226  mdfd->mdfd_vfd = fd;
227  mdfd->mdfd_segno = 0;
228 }
229 
230 /*
231  * mdunlink() -- Unlink a relation.
232  *
233  * Note that we're passed a RelFileNodeBackend --- by the time this is called,
234  * there won't be an SMgrRelation hashtable entry anymore.
235  *
236  * forkNum can be a fork number to delete a specific fork, or InvalidForkNumber
237  * to delete all forks.
238  *
239  * For regular relations, we don't unlink the first segment file of the rel,
240  * but just truncate it to zero length, and record a request to unlink it after
241  * the next checkpoint. Additional segments can be unlinked immediately,
242  * however. Leaving the empty file in place prevents that relfilenode
243  * number from being reused. The scenario this protects us from is:
244  * 1. We delete a relation (and commit, and actually remove its file).
245  * 2. We create a new relation, which by chance gets the same relfilenode as
246  * the just-deleted one (OIDs must've wrapped around for that to happen).
247  * 3. We crash before another checkpoint occurs.
248  * During replay, we would delete the file and then recreate it, which is fine
249  * if the contents of the file were repopulated by subsequent WAL entries.
250  * But if we didn't WAL-log insertions, but instead relied on fsyncing the
251  * file after populating it (as we do at wal_level=minimal), the contents of
252  * the file would be lost forever. By leaving the empty file until after the
253  * next checkpoint, we prevent reassignment of the relfilenode number until
254  * it's safe, because relfilenode assignment skips over any existing file.
255  *
256  * We do not need to go through this dance for temp relations, though, because
257  * we never make WAL entries for temp rels, and so a temp rel poses no threat
258  * to the health of a regular rel that has taken over its relfilenode number.
259  * The fact that temp rels and regular rels have different file naming
260  * patterns provides additional safety.
261  *
262  * All the above applies only to the relation's main fork; other forks can
263  * just be removed immediately, since they are not needed to prevent the
264  * relfilenode number from being recycled. Also, we do not carefully
265  * track whether other forks have been created or not, but just attempt to
266  * unlink them unconditionally; so we should never complain about ENOENT.
267  *
268  * If isRedo is true, it's unsurprising for the relation to be already gone.
269  * Also, we should remove the file immediately instead of queuing a request
270  * for later, since during redo there's no possibility of creating a
271  * conflicting relation.
272  *
273  * Note: any failure should be reported as WARNING not ERROR, because
274  * we are usually not in a transaction anymore when this is called.
275  */
276 void
277 mdunlink(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
278 {
279  /* Now do the per-fork work */
280  if (forkNum == InvalidForkNumber)
281  {
282  for (forkNum = 0; forkNum <= MAX_FORKNUM; forkNum++)
283  mdunlinkfork(rnode, forkNum, isRedo);
284  }
285  else
286  mdunlinkfork(rnode, forkNum, isRedo);
287 }
288 
289 /*
290  * Truncate a file to release disk space.
291  */
292 static int
293 do_truncate(const char *path)
294 {
295  int save_errno;
296  int ret;
297 
298  ret = pg_truncate(path, 0);
299 
300  /* Log a warning here to avoid repetition in callers. */
301  if (ret < 0 && errno != ENOENT)
302  {
303  save_errno = errno;
306  errmsg("could not truncate file \"%s\": %m", path)));
307  errno = save_errno;
308  }
309 
310  return ret;
311 }
312 
313 static void
314 mdunlinkfork(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
315 {
316  char *path;
317  int ret;
318 
319  path = relpath(rnode, forkNum);
320 
321  /*
322  * Delete or truncate the first segment.
323  */
324  if (isRedo || forkNum != MAIN_FORKNUM || RelFileNodeBackendIsTemp(rnode))
325  {
326  if (!RelFileNodeBackendIsTemp(rnode))
327  {
328  /* Prevent other backends' fds from holding on to the disk space */
329  ret = do_truncate(path);
330 
331  /* Forget any pending sync requests for the first segment */
332  register_forget_request(rnode, forkNum, 0 /* first seg */ );
333  }
334  else
335  ret = 0;
336 
337  /* Next unlink the file, unless it was already found to be missing */
338  if (ret == 0 || errno != ENOENT)
339  {
340  ret = unlink(path);
341  if (ret < 0 && errno != ENOENT)
344  errmsg("could not remove file \"%s\": %m", path)));
345  }
346  }
347  else
348  {
349  /* Prevent other backends' fds from holding on to the disk space */
350  ret = do_truncate(path);
351 
352  /* Register request to unlink first segment later */
353  register_unlink_segment(rnode, forkNum, 0 /* first seg */ );
354  }
355 
356  /*
357  * Delete any additional segments.
358  */
359  if (ret >= 0)
360  {
361  char *segpath = (char *) palloc(strlen(path) + 12);
362  BlockNumber segno;
363 
364  /*
365  * Note that because we loop until getting ENOENT, we will correctly
366  * remove all inactive segments as well as active ones.
367  */
368  for (segno = 1;; segno++)
369  {
370  sprintf(segpath, "%s.%u", path, segno);
371 
372  if (!RelFileNodeBackendIsTemp(rnode))
373  {
374  /*
375  * Prevent other backends' fds from holding on to the disk
376  * space.
377  */
378  if (do_truncate(segpath) < 0 && errno == ENOENT)
379  break;
380 
381  /*
382  * Forget any pending sync requests for this segment before we
383  * try to unlink.
384  */
385  register_forget_request(rnode, forkNum, segno);
386  }
387 
388  if (unlink(segpath) < 0)
389  {
390  /* ENOENT is expected after the last segment... */
391  if (errno != ENOENT)
394  errmsg("could not remove file \"%s\": %m", segpath)));
395  break;
396  }
397  }
398  pfree(segpath);
399  }
400 
401  pfree(path);
402 }
403 
404 /*
405  * mdextend() -- Add a block to the specified relation.
406  *
407  * The semantics are nearly the same as mdwrite(): write at the
408  * specified position. However, this is to be used for the case of
409  * extending a relation (i.e., blocknum is at or beyond the current
410  * EOF). Note that we assume writing a block beyond current EOF
411  * causes intervening file space to become filled with zeroes.
412  */
413 void
414 mdextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
415  char *buffer, bool skipFsync)
416 {
417  off_t seekpos;
418  int nbytes;
419  MdfdVec *v;
420 
421  /* This assert is too expensive to have on normally ... */
422 #ifdef CHECK_WRITE_VS_EXTEND
423  Assert(blocknum >= mdnblocks(reln, forknum));
424 #endif
425 
426  /*
427  * If a relation manages to grow to 2^32-1 blocks, refuse to extend it any
428  * more --- we mustn't create a block whose number actually is
429  * InvalidBlockNumber.
430  */
431  if (blocknum == InvalidBlockNumber)
432  ereport(ERROR,
433  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
434  errmsg("cannot extend file \"%s\" beyond %u blocks",
435  relpath(reln->smgr_rnode, forknum),
437 
438  v = _mdfd_getseg(reln, forknum, blocknum, skipFsync, EXTENSION_CREATE);
439 
440  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
441 
442  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
443 
444  if ((nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_EXTEND)) != BLCKSZ)
445  {
446  if (nbytes < 0)
447  ereport(ERROR,
449  errmsg("could not extend file \"%s\": %m",
450  FilePathName(v->mdfd_vfd)),
451  errhint("Check free disk space.")));
452  /* short write: complain appropriately */
453  ereport(ERROR,
454  (errcode(ERRCODE_DISK_FULL),
455  errmsg("could not extend file \"%s\": wrote only %d of %d bytes at block %u",
457  nbytes, BLCKSZ, blocknum),
458  errhint("Check free disk space.")));
459  }
460 
461  if (!skipFsync && !SmgrIsTemp(reln))
462  register_dirty_segment(reln, forknum, v);
463 
464  Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
465 }
466 
467 /*
468  * mdopenfork() -- Open one fork of the specified relation.
469  *
470  * Note we only open the first segment, when there are multiple segments.
471  *
472  * If first segment is not present, either ereport or return NULL according
473  * to "behavior". We treat EXTENSION_CREATE the same as EXTENSION_FAIL;
474  * EXTENSION_CREATE means it's OK to extend an existing relation, not to
475  * invent one out of whole cloth.
476  */
477 static MdfdVec *
478 mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
479 {
480  MdfdVec *mdfd;
481  char *path;
482  File fd;
483 
484  /* No work if already open */
485  if (reln->md_num_open_segs[forknum] > 0)
486  return &reln->md_seg_fds[forknum][0];
487 
488  path = relpath(reln->smgr_rnode, forknum);
489 
490  fd = PathNameOpenFile(path, O_RDWR | PG_BINARY);
491 
492  if (fd < 0)
493  {
494  if ((behavior & EXTENSION_RETURN_NULL) &&
495  FILE_POSSIBLY_DELETED(errno))
496  {
497  pfree(path);
498  return NULL;
499  }
500  ereport(ERROR,
502  errmsg("could not open file \"%s\": %m", path)));
503  }
504 
505  pfree(path);
506 
507  _fdvec_resize(reln, forknum, 1);
508  mdfd = &reln->md_seg_fds[forknum][0];
509  mdfd->mdfd_vfd = fd;
510  mdfd->mdfd_segno = 0;
511 
512  Assert(_mdnblocks(reln, forknum, mdfd) <= ((BlockNumber) RELSEG_SIZE));
513 
514  return mdfd;
515 }
516 
517 /*
518  * mdopen() -- Initialize newly-opened relation.
519  */
520 void
522 {
523  /* mark it not open */
524  for (int forknum = 0; forknum <= MAX_FORKNUM; forknum++)
525  reln->md_num_open_segs[forknum] = 0;
526 }
527 
528 /*
529  * mdclose() -- Close the specified relation, if it isn't closed already.
530  */
531 void
533 {
534  int nopensegs = reln->md_num_open_segs[forknum];
535 
536  /* No work if already closed */
537  if (nopensegs == 0)
538  return;
539 
540  /* close segments starting from the end */
541  while (nopensegs > 0)
542  {
543  MdfdVec *v = &reln->md_seg_fds[forknum][nopensegs - 1];
544 
545  FileClose(v->mdfd_vfd);
546  _fdvec_resize(reln, forknum, nopensegs - 1);
547  nopensegs--;
548  }
549 }
550 
551 /*
552  * mdprefetch() -- Initiate asynchronous read of the specified block of a relation
553  */
554 bool
556 {
557 #ifdef USE_PREFETCH
558  off_t seekpos;
559  MdfdVec *v;
560 
561  v = _mdfd_getseg(reln, forknum, blocknum, false,
563  if (v == NULL)
564  return false;
565 
566  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
567 
568  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
569 
570  (void) FilePrefetch(v->mdfd_vfd, seekpos, BLCKSZ, WAIT_EVENT_DATA_FILE_PREFETCH);
571 #endif /* USE_PREFETCH */
572 
573  return true;
574 }
575 
576 /*
577  * mdwriteback() -- Tell the kernel to write pages back to storage.
578  *
579  * This accepts a range of blocks because flushing several pages at once is
580  * considerably more efficient than doing so individually.
581  */
582 void
584  BlockNumber blocknum, BlockNumber nblocks)
585 {
586  /*
587  * Issue flush requests in as few requests as possible; have to split at
588  * segment boundaries though, since those are actually separate files.
589  */
590  while (nblocks > 0)
591  {
592  BlockNumber nflush = nblocks;
593  off_t seekpos;
594  MdfdVec *v;
595  int segnum_start,
596  segnum_end;
597 
598  v = _mdfd_getseg(reln, forknum, blocknum, true /* not used */ ,
600 
601  /*
602  * We might be flushing buffers of already removed relations, that's
603  * ok, just ignore that case.
604  */
605  if (!v)
606  return;
607 
608  /* compute offset inside the current segment */
609  segnum_start = blocknum / RELSEG_SIZE;
610 
611  /* compute number of desired writes within the current segment */
612  segnum_end = (blocknum + nblocks - 1) / RELSEG_SIZE;
613  if (segnum_start != segnum_end)
614  nflush = RELSEG_SIZE - (blocknum % ((BlockNumber) RELSEG_SIZE));
615 
616  Assert(nflush >= 1);
617  Assert(nflush <= nblocks);
618 
619  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
620 
621  FileWriteback(v->mdfd_vfd, seekpos, (off_t) BLCKSZ * nflush, WAIT_EVENT_DATA_FILE_FLUSH);
622 
623  nblocks -= nflush;
624  blocknum += nflush;
625  }
626 }
627 
628 /*
629  * mdread() -- Read the specified block from a relation.
630  */
631 void
632 mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
633  char *buffer)
634 {
635  off_t seekpos;
636  int nbytes;
637  MdfdVec *v;
638 
639  TRACE_POSTGRESQL_SMGR_MD_READ_START(forknum, blocknum,
640  reln->smgr_rnode.node.spcNode,
641  reln->smgr_rnode.node.dbNode,
642  reln->smgr_rnode.node.relNode,
643  reln->smgr_rnode.backend);
644 
645  v = _mdfd_getseg(reln, forknum, blocknum, false,
647 
648  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
649 
650  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
651 
652  nbytes = FileRead(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_READ);
653 
654  TRACE_POSTGRESQL_SMGR_MD_READ_DONE(forknum, blocknum,
655  reln->smgr_rnode.node.spcNode,
656  reln->smgr_rnode.node.dbNode,
657  reln->smgr_rnode.node.relNode,
658  reln->smgr_rnode.backend,
659  nbytes,
660  BLCKSZ);
661 
662  if (nbytes != BLCKSZ)
663  {
664  if (nbytes < 0)
665  ereport(ERROR,
667  errmsg("could not read block %u in file \"%s\": %m",
668  blocknum, FilePathName(v->mdfd_vfd))));
669 
670  /*
671  * Short read: we are at or past EOF, or we read a partial block at
672  * EOF. Normally this is an error; upper levels should never try to
673  * read a nonexistent block. However, if zero_damaged_pages is ON or
674  * we are InRecovery, we should instead return zeroes without
675  * complaining. This allows, for example, the case of trying to
676  * update a block that was later truncated away.
677  */
679  MemSet(buffer, 0, BLCKSZ);
680  else
681  ereport(ERROR,
683  errmsg("could not read block %u in file \"%s\": read only %d of %d bytes",
684  blocknum, FilePathName(v->mdfd_vfd),
685  nbytes, BLCKSZ)));
686  }
687 }
688 
689 /*
690  * mdwrite() -- Write the supplied block at the appropriate location.
691  *
692  * This is to be used only for updating already-existing blocks of a
693  * relation (ie, those before the current EOF). To extend a relation,
694  * use mdextend().
695  */
696 void
697 mdwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
698  char *buffer, bool skipFsync)
699 {
700  off_t seekpos;
701  int nbytes;
702  MdfdVec *v;
703 
704  /* This assert is too expensive to have on normally ... */
705 #ifdef CHECK_WRITE_VS_EXTEND
706  Assert(blocknum < mdnblocks(reln, forknum));
707 #endif
708 
709  TRACE_POSTGRESQL_SMGR_MD_WRITE_START(forknum, blocknum,
710  reln->smgr_rnode.node.spcNode,
711  reln->smgr_rnode.node.dbNode,
712  reln->smgr_rnode.node.relNode,
713  reln->smgr_rnode.backend);
714 
715  v = _mdfd_getseg(reln, forknum, blocknum, skipFsync,
717 
718  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
719 
720  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
721 
722  nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_WRITE);
723 
724  TRACE_POSTGRESQL_SMGR_MD_WRITE_DONE(forknum, blocknum,
725  reln->smgr_rnode.node.spcNode,
726  reln->smgr_rnode.node.dbNode,
727  reln->smgr_rnode.node.relNode,
728  reln->smgr_rnode.backend,
729  nbytes,
730  BLCKSZ);
731 
732  if (nbytes != BLCKSZ)
733  {
734  if (nbytes < 0)
735  ereport(ERROR,
737  errmsg("could not write block %u in file \"%s\": %m",
738  blocknum, FilePathName(v->mdfd_vfd))));
739  /* short write: complain appropriately */
740  ereport(ERROR,
741  (errcode(ERRCODE_DISK_FULL),
742  errmsg("could not write block %u in file \"%s\": wrote only %d of %d bytes",
743  blocknum,
745  nbytes, BLCKSZ),
746  errhint("Check free disk space.")));
747  }
748 
749  if (!skipFsync && !SmgrIsTemp(reln))
750  register_dirty_segment(reln, forknum, v);
751 }
752 
753 /*
754  * mdnblocks() -- Get the number of blocks stored in a relation.
755  *
756  * Important side effect: all active segments of the relation are opened
757  * and added to the md_seg_fds array. If this routine has not been
758  * called, then only segments up to the last one actually touched
759  * are present in the array.
760  */
763 {
764  MdfdVec *v;
765  BlockNumber nblocks;
766  BlockNumber segno;
767 
768  mdopenfork(reln, forknum, EXTENSION_FAIL);
769 
770  /* mdopen has opened the first segment */
771  Assert(reln->md_num_open_segs[forknum] > 0);
772 
773  /*
774  * Start from the last open segments, to avoid redundant seeks. We have
775  * previously verified that these segments are exactly RELSEG_SIZE long,
776  * and it's useless to recheck that each time.
777  *
778  * NOTE: this assumption could only be wrong if another backend has
779  * truncated the relation. We rely on higher code levels to handle that
780  * scenario by closing and re-opening the md fd, which is handled via
781  * relcache flush. (Since the checkpointer doesn't participate in
782  * relcache flush, it could have segment entries for inactive segments;
783  * that's OK because the checkpointer never needs to compute relation
784  * size.)
785  */
786  segno = reln->md_num_open_segs[forknum] - 1;
787  v = &reln->md_seg_fds[forknum][segno];
788 
789  for (;;)
790  {
791  nblocks = _mdnblocks(reln, forknum, v);
792  if (nblocks > ((BlockNumber) RELSEG_SIZE))
793  elog(FATAL, "segment too big");
794  if (nblocks < ((BlockNumber) RELSEG_SIZE))
795  return (segno * ((BlockNumber) RELSEG_SIZE)) + nblocks;
796 
797  /*
798  * If segment is exactly RELSEG_SIZE, advance to next one.
799  */
800  segno++;
801 
802  /*
803  * We used to pass O_CREAT here, but that has the disadvantage that it
804  * might create a segment which has vanished through some operating
805  * system misadventure. In such a case, creating the segment here
806  * undermines _mdfd_getseg's attempts to notice and report an error
807  * upon access to a missing segment.
808  */
809  v = _mdfd_openseg(reln, forknum, segno, 0);
810  if (v == NULL)
811  return segno * ((BlockNumber) RELSEG_SIZE);
812  }
813 }
814 
815 /*
816  * mdtruncate() -- Truncate relation to specified number of blocks.
817  */
818 void
820 {
821  BlockNumber curnblk;
822  BlockNumber priorblocks;
823  int curopensegs;
824 
825  /*
826  * NOTE: mdnblocks makes sure we have opened all active segments, so that
827  * truncation loop will get them all!
828  */
829  curnblk = mdnblocks(reln, forknum);
830  if (nblocks > curnblk)
831  {
832  /* Bogus request ... but no complaint if InRecovery */
833  if (InRecovery)
834  return;
835  ereport(ERROR,
836  (errmsg("could not truncate file \"%s\" to %u blocks: it's only %u blocks now",
837  relpath(reln->smgr_rnode, forknum),
838  nblocks, curnblk)));
839  }
840  if (nblocks == curnblk)
841  return; /* no work */
842 
843  /*
844  * Truncate segments, starting at the last one. Starting at the end makes
845  * managing the memory for the fd array easier, should there be errors.
846  */
847  curopensegs = reln->md_num_open_segs[forknum];
848  while (curopensegs > 0)
849  {
850  MdfdVec *v;
851 
852  priorblocks = (curopensegs - 1) * RELSEG_SIZE;
853 
854  v = &reln->md_seg_fds[forknum][curopensegs - 1];
855 
856  if (priorblocks > nblocks)
857  {
858  /*
859  * This segment is no longer active. We truncate the file, but do
860  * not delete it, for reasons explained in the header comments.
861  */
863  ereport(ERROR,
865  errmsg("could not truncate file \"%s\": %m",
866  FilePathName(v->mdfd_vfd))));
867 
868  if (!SmgrIsTemp(reln))
869  register_dirty_segment(reln, forknum, v);
870 
871  /* we never drop the 1st segment */
872  Assert(v != &reln->md_seg_fds[forknum][0]);
873 
874  FileClose(v->mdfd_vfd);
875  _fdvec_resize(reln, forknum, curopensegs - 1);
876  }
877  else if (priorblocks + ((BlockNumber) RELSEG_SIZE) > nblocks)
878  {
879  /*
880  * This is the last segment we want to keep. Truncate the file to
881  * the right length. NOTE: if nblocks is exactly a multiple K of
882  * RELSEG_SIZE, we will truncate the K+1st segment to 0 length but
883  * keep it. This adheres to the invariant given in the header
884  * comments.
885  */
886  BlockNumber lastsegblocks = nblocks - priorblocks;
887 
888  if (FileTruncate(v->mdfd_vfd, (off_t) lastsegblocks * BLCKSZ, WAIT_EVENT_DATA_FILE_TRUNCATE) < 0)
889  ereport(ERROR,
891  errmsg("could not truncate file \"%s\" to %u blocks: %m",
893  nblocks)));
894  if (!SmgrIsTemp(reln))
895  register_dirty_segment(reln, forknum, v);
896  }
897  else
898  {
899  /*
900  * We still need this segment, so nothing to do for this and any
901  * earlier segment.
902  */
903  break;
904  }
905  curopensegs--;
906  }
907 }
908 
909 /*
910  * mdimmedsync() -- Immediately sync a relation to stable storage.
911  *
912  * Note that only writes already issued are synced; this routine knows
913  * nothing of dirty buffers that may exist inside the buffer manager. We
914  * sync active and inactive segments; smgrDoPendingSyncs() relies on this.
915  * Consider a relation skipping WAL. Suppose a checkpoint syncs blocks of
916  * some segment, then mdtruncate() renders that segment inactive. If we
917  * crash before the next checkpoint syncs the newly-inactive segment, that
918  * segment may survive recovery, reintroducing unwanted data into the table.
919  */
920 void
922 {
923  int segno;
924  int min_inactive_seg;
925 
926  /*
927  * NOTE: mdnblocks makes sure we have opened all active segments, so that
928  * fsync loop will get them all!
929  */
930  mdnblocks(reln, forknum);
931 
932  min_inactive_seg = segno = reln->md_num_open_segs[forknum];
933 
934  /*
935  * Temporarily open inactive segments, then close them after sync. There
936  * may be some inactive segments left opened after fsync() error, but that
937  * is harmless. We don't bother to clean them up and take a risk of
938  * further trouble. The next mdclose() will soon close them.
939  */
940  while (_mdfd_openseg(reln, forknum, segno, 0) != NULL)
941  segno++;
942 
943  while (segno > 0)
944  {
945  MdfdVec *v = &reln->md_seg_fds[forknum][segno - 1];
946 
950  errmsg("could not fsync file \"%s\": %m",
951  FilePathName(v->mdfd_vfd))));
952 
953  /* Close inactive segments immediately */
954  if (segno > min_inactive_seg)
955  {
956  FileClose(v->mdfd_vfd);
957  _fdvec_resize(reln, forknum, segno - 1);
958  }
959 
960  segno--;
961  }
962 }
963 
964 /*
965  * register_dirty_segment() -- Mark a relation segment as needing fsync
966  *
967  * If there is a local pending-ops table, just make an entry in it for
968  * ProcessSyncRequests to process later. Otherwise, try to pass off the
969  * fsync request to the checkpointer process. If that fails, just do the
970  * fsync locally before returning (we hope this will not happen often
971  * enough to be a performance problem).
972  */
973 static void
975 {
976  FileTag tag;
977 
978  INIT_MD_FILETAG(tag, reln->smgr_rnode.node, forknum, seg->mdfd_segno);
979 
980  /* Temp relations should never be fsync'd */
981  Assert(!SmgrIsTemp(reln));
982 
983  if (!RegisterSyncRequest(&tag, SYNC_REQUEST, false /* retryOnError */ ))
984  {
985  ereport(DEBUG1,
986  (errmsg("could not forward fsync request because request queue is full")));
987 
991  errmsg("could not fsync file \"%s\": %m",
992  FilePathName(seg->mdfd_vfd))));
993  }
994 }
995 
996 /*
997  * register_unlink_segment() -- Schedule a file to be deleted after next checkpoint
998  */
999 static void
1001  BlockNumber segno)
1002 {
1003  FileTag tag;
1004 
1005  INIT_MD_FILETAG(tag, rnode.node, forknum, segno);
1006 
1007  /* Should never be used with temp relations */
1009 
1010  RegisterSyncRequest(&tag, SYNC_UNLINK_REQUEST, true /* retryOnError */ );
1011 }
1012 
1013 /*
1014  * register_forget_request() -- forget any fsyncs for a relation fork's segment
1015  */
1016 static void
1018  BlockNumber segno)
1019 {
1020  FileTag tag;
1021 
1022  INIT_MD_FILETAG(tag, rnode.node, forknum, segno);
1023 
1024  RegisterSyncRequest(&tag, SYNC_FORGET_REQUEST, true /* retryOnError */ );
1025 }
1026 
1027 /*
1028  * ForgetDatabaseSyncRequests -- forget any fsyncs and unlinks for a DB
1029  */
1030 void
1032 {
1033  FileTag tag;
1034  RelFileNode rnode;
1035 
1036  rnode.dbNode = dbid;
1037  rnode.spcNode = 0;
1038  rnode.relNode = 0;
1039 
1041 
1042  RegisterSyncRequest(&tag, SYNC_FILTER_REQUEST, true /* retryOnError */ );
1043 }
1044 
1045 /*
1046  * DropRelationFiles -- drop files of all given relations
1047  */
1048 void
1049 DropRelationFiles(RelFileNode *delrels, int ndelrels, bool isRedo)
1050 {
1051  SMgrRelation *srels;
1052  int i;
1053 
1054  srels = palloc(sizeof(SMgrRelation) * ndelrels);
1055  for (i = 0; i < ndelrels; i++)
1056  {
1057  SMgrRelation srel = smgropen(delrels[i], InvalidBackendId);
1058 
1059  if (isRedo)
1060  {
1061  ForkNumber fork;
1062 
1063  for (fork = 0; fork <= MAX_FORKNUM; fork++)
1064  XLogDropRelation(delrels[i], fork);
1065  }
1066  srels[i] = srel;
1067  }
1068 
1069  smgrdounlinkall(srels, ndelrels, isRedo);
1070 
1071  for (i = 0; i < ndelrels; i++)
1072  smgrclose(srels[i]);
1073  pfree(srels);
1074 }
1075 
1076 
1077 /*
1078  * _fdvec_resize() -- Resize the fork's open segments array
1079  */
1080 static void
1082  ForkNumber forknum,
1083  int nseg)
1084 {
1085  if (nseg == 0)
1086  {
1087  if (reln->md_num_open_segs[forknum] > 0)
1088  {
1089  pfree(reln->md_seg_fds[forknum]);
1090  reln->md_seg_fds[forknum] = NULL;
1091  }
1092  }
1093  else if (reln->md_num_open_segs[forknum] == 0)
1094  {
1095  reln->md_seg_fds[forknum] =
1096  MemoryContextAlloc(MdCxt, sizeof(MdfdVec) * nseg);
1097  }
1098  else
1099  {
1100  /*
1101  * It doesn't seem worthwhile complicating the code to amortize
1102  * repalloc() calls. Those are far faster than PathNameOpenFile() or
1103  * FileClose(), and the memory context internally will sometimes avoid
1104  * doing an actual reallocation.
1105  */
1106  reln->md_seg_fds[forknum] =
1107  repalloc(reln->md_seg_fds[forknum],
1108  sizeof(MdfdVec) * nseg);
1109  }
1110 
1111  reln->md_num_open_segs[forknum] = nseg;
1112 }
1113 
1114 /*
1115  * Return the filename for the specified segment of the relation. The
1116  * returned string is palloc'd.
1117  */
1118 static char *
1120 {
1121  char *path,
1122  *fullpath;
1123 
1124  path = relpath(reln->smgr_rnode, forknum);
1125 
1126  if (segno > 0)
1127  {
1128  fullpath = psprintf("%s.%u", path, segno);
1129  pfree(path);
1130  }
1131  else
1132  fullpath = path;
1133 
1134  return fullpath;
1135 }
1136 
1137 /*
1138  * Open the specified segment of the relation,
1139  * and make a MdfdVec object for it. Returns NULL on failure.
1140  */
1141 static MdfdVec *
1143  int oflags)
1144 {
1145  MdfdVec *v;
1146  File fd;
1147  char *fullpath;
1148 
1149  fullpath = _mdfd_segpath(reln, forknum, segno);
1150 
1151  /* open the file */
1152  fd = PathNameOpenFile(fullpath, O_RDWR | PG_BINARY | oflags);
1153 
1154  pfree(fullpath);
1155 
1156  if (fd < 0)
1157  return NULL;
1158 
1159  /*
1160  * Segments are always opened in order from lowest to highest, so we must
1161  * be adding a new one at the end.
1162  */
1163  Assert(segno == reln->md_num_open_segs[forknum]);
1164 
1165  _fdvec_resize(reln, forknum, segno + 1);
1166 
1167  /* fill the entry */
1168  v = &reln->md_seg_fds[forknum][segno];
1169  v->mdfd_vfd = fd;
1170  v->mdfd_segno = segno;
1171 
1172  Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
1173 
1174  /* all done */
1175  return v;
1176 }
1177 
1178 /*
1179  * _mdfd_getseg() -- Find the segment of the relation holding the
1180  * specified block.
1181  *
1182  * If the segment doesn't exist, we ereport, return NULL, or create the
1183  * segment, according to "behavior". Note: skipFsync is only used in the
1184  * EXTENSION_CREATE case.
1185  */
1186 static MdfdVec *
1188  bool skipFsync, int behavior)
1189 {
1190  MdfdVec *v;
1191  BlockNumber targetseg;
1192  BlockNumber nextsegno;
1193 
1194  /* some way to handle non-existent segments needs to be specified */
1195  Assert(behavior &
1197 
1198  targetseg = blkno / ((BlockNumber) RELSEG_SIZE);
1199 
1200  /* if an existing and opened segment, we're done */
1201  if (targetseg < reln->md_num_open_segs[forknum])
1202  {
1203  v = &reln->md_seg_fds[forknum][targetseg];
1204  return v;
1205  }
1206 
1207  /*
1208  * The target segment is not yet open. Iterate over all the segments
1209  * between the last opened and the target segment. This way missing
1210  * segments either raise an error, or get created (according to
1211  * 'behavior'). Start with either the last opened, or the first segment if
1212  * none was opened before.
1213  */
1214  if (reln->md_num_open_segs[forknum] > 0)
1215  v = &reln->md_seg_fds[forknum][reln->md_num_open_segs[forknum] - 1];
1216  else
1217  {
1218  v = mdopenfork(reln, forknum, behavior);
1219  if (!v)
1220  return NULL; /* if behavior & EXTENSION_RETURN_NULL */
1221  }
1222 
1223  for (nextsegno = reln->md_num_open_segs[forknum];
1224  nextsegno <= targetseg; nextsegno++)
1225  {
1226  BlockNumber nblocks = _mdnblocks(reln, forknum, v);
1227  int flags = 0;
1228 
1229  Assert(nextsegno == v->mdfd_segno + 1);
1230 
1231  if (nblocks > ((BlockNumber) RELSEG_SIZE))
1232  elog(FATAL, "segment too big");
1233 
1234  if ((behavior & EXTENSION_CREATE) ||
1235  (InRecovery && (behavior & EXTENSION_CREATE_RECOVERY)))
1236  {
1237  /*
1238  * Normally we will create new segments only if authorized by the
1239  * caller (i.e., we are doing mdextend()). But when doing WAL
1240  * recovery, create segments anyway; this allows cases such as
1241  * replaying WAL data that has a write into a high-numbered
1242  * segment of a relation that was later deleted. We want to go
1243  * ahead and create the segments so we can finish out the replay.
1244  *
1245  * We have to maintain the invariant that segments before the last
1246  * active segment are of size RELSEG_SIZE; therefore, if
1247  * extending, pad them out with zeroes if needed. (This only
1248  * matters if in recovery, or if the caller is extending the
1249  * relation discontiguously, but that can happen in hash indexes.)
1250  */
1251  if (nblocks < ((BlockNumber) RELSEG_SIZE))
1252  {
1253  char *zerobuf = palloc0(BLCKSZ);
1254 
1255  mdextend(reln, forknum,
1256  nextsegno * ((BlockNumber) RELSEG_SIZE) - 1,
1257  zerobuf, skipFsync);
1258  pfree(zerobuf);
1259  }
1260  flags = O_CREAT;
1261  }
1262  else if (!(behavior & EXTENSION_DONT_CHECK_SIZE) &&
1263  nblocks < ((BlockNumber) RELSEG_SIZE))
1264  {
1265  /*
1266  * When not extending (or explicitly including truncated
1267  * segments), only open the next segment if the current one is
1268  * exactly RELSEG_SIZE. If not (this branch), either return NULL
1269  * or fail.
1270  */
1271  if (behavior & EXTENSION_RETURN_NULL)
1272  {
1273  /*
1274  * Some callers discern between reasons for _mdfd_getseg()
1275  * returning NULL based on errno. As there's no failing
1276  * syscall involved in this case, explicitly set errno to
1277  * ENOENT, as that seems the closest interpretation.
1278  */
1279  errno = ENOENT;
1280  return NULL;
1281  }
1282 
1283  ereport(ERROR,
1285  errmsg("could not open file \"%s\" (target block %u): previous segment is only %u blocks",
1286  _mdfd_segpath(reln, forknum, nextsegno),
1287  blkno, nblocks)));
1288  }
1289 
1290  v = _mdfd_openseg(reln, forknum, nextsegno, flags);
1291 
1292  if (v == NULL)
1293  {
1294  if ((behavior & EXTENSION_RETURN_NULL) &&
1295  FILE_POSSIBLY_DELETED(errno))
1296  return NULL;
1297  ereport(ERROR,
1299  errmsg("could not open file \"%s\" (target block %u): %m",
1300  _mdfd_segpath(reln, forknum, nextsegno),
1301  blkno)));
1302  }
1303  }
1304 
1305  return v;
1306 }
1307 
1308 /*
1309  * Get number of blocks present in a single disk file
1310  */
1311 static BlockNumber
1313 {
1314  off_t len;
1315 
1316  len = FileSize(seg->mdfd_vfd);
1317  if (len < 0)
1318  ereport(ERROR,
1320  errmsg("could not seek to end of file \"%s\": %m",
1321  FilePathName(seg->mdfd_vfd))));
1322  /* note that this calculation will ignore any partial block at EOF */
1323  return (BlockNumber) (len / BLCKSZ);
1324 }
1325 
1326 /*
1327  * Sync a file to disk, given a file tag. Write the path into an output
1328  * buffer so the caller can use it in error messages.
1329  *
1330  * Return 0 on success, -1 on failure, with errno set.
1331  */
1332 int
1333 mdsyncfiletag(const FileTag *ftag, char *path)
1334 {
1336  File file;
1337  bool need_to_close;
1338  int result,
1339  save_errno;
1340 
1341  /* See if we already have the file open, or need to open it. */
1342  if (ftag->segno < reln->md_num_open_segs[ftag->forknum])
1343  {
1344  file = reln->md_seg_fds[ftag->forknum][ftag->segno].mdfd_vfd;
1345  strlcpy(path, FilePathName(file), MAXPGPATH);
1346  need_to_close = false;
1347  }
1348  else
1349  {
1350  char *p;
1351 
1352  p = _mdfd_segpath(reln, ftag->forknum, ftag->segno);
1353  strlcpy(path, p, MAXPGPATH);
1354  pfree(p);
1355 
1356  file = PathNameOpenFile(path, O_RDWR | PG_BINARY);
1357  if (file < 0)
1358  return -1;
1359  need_to_close = true;
1360  }
1361 
1362  /* Sync the file. */
1363  result = FileSync(file, WAIT_EVENT_DATA_FILE_SYNC);
1364  save_errno = errno;
1365 
1366  if (need_to_close)
1367  FileClose(file);
1368 
1369  errno = save_errno;
1370  return result;
1371 }
1372 
1373 /*
1374  * Unlink a file, given a file tag. Write the path into an output
1375  * buffer so the caller can use it in error messages.
1376  *
1377  * Return 0 on success, -1 on failure, with errno set.
1378  */
1379 int
1380 mdunlinkfiletag(const FileTag *ftag, char *path)
1381 {
1382  char *p;
1383 
1384  /* Compute the path. */
1385  p = relpathperm(ftag->rnode, MAIN_FORKNUM);
1386  strlcpy(path, p, MAXPGPATH);
1387  pfree(p);
1388 
1389  /* Try to unlink the file. */
1390  return unlink(path);
1391 }
1392 
1393 /*
1394  * Check if a given candidate request matches a given tag, when processing
1395  * a SYNC_FILTER_REQUEST request. This will be called for all pending
1396  * requests to find out whether to forget them.
1397  */
1398 bool
1399 mdfiletagmatches(const FileTag *ftag, const FileTag *candidate)
1400 {
1401  /*
1402  * For now we only use filter requests as a way to drop all scheduled
1403  * callbacks relating to a given database, when dropping the database.
1404  * We'll return true for all candidates that have the same database OID as
1405  * the ftag from the SYNC_FILTER_REQUEST request, so they're forgotten.
1406  */
1407  return ftag->rnode.dbNode == candidate->rnode.dbNode;
1408 }
uint32 segno
Definition: sync.h:55
#define relpathperm(rnode, forknum)
Definition: relpath.h:83
void mdimmedsync(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:921
#define AllocSetContextCreate
Definition: memutils.h:170
void smgrclose(SMgrRelation reln)
Definition: smgr.c:257
#define DEBUG1
Definition: elog.h:25
static MdfdVec * _mdfd_getseg(SMgrRelation reln, ForkNumber forkno, BlockNumber blkno, bool skipFsync, int behavior)
Definition: md.c:1187
int errhint(const char *fmt,...)
Definition: elog.c:1149
#define EXTENSION_DONT_CHECK_SIZE
Definition: md.c:118
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1464
#define RelFileNodeBackendIsTemp(rnode)
Definition: relfilenode.h:78
BlockNumber mdnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:762
int pg_truncate(const char *path, off_t length)
Definition: fd.c:629
char * psprintf(const char *fmt,...)
Definition: psprintf.c:46
BlockNumber mdfd_segno
Definition: md.c:85
bool InRecovery
Definition: xlog.c:205
struct _MdfdVec MdfdVec
off_t FileSize(File file)
Definition: fd.c:2175
static int do_truncate(const char *path)
Definition: md.c:293
int16 forknum
Definition: sync.h:53
int errcode(int sqlerrcode)
Definition: elog.c:691
void smgrdounlinkall(SMgrRelation *rels, int nrels, bool isRedo)
Definition: smgr.c:385
#define MemSet(start, val, len)
Definition: c.h:1004
RelFileNode rnode
Definition: sync.h:54
uint32 BlockNumber
Definition: block.h:31
#define INIT_MD_FILETAG(a, xx_rnode, xx_forknum, xx_segno)
Definition: md.c:92
unsigned int Oid
Definition: postgres_ext.h:31
void mdinit(void)
Definition: md.c:148
char * FilePathName(File file)
Definition: fd.c:2227
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1267
#define EXTENSION_FAIL
Definition: md.c:104
static void register_forget_request(RelFileNodeBackend rnode, ForkNumber forknum, BlockNumber segno)
Definition: md.c:1017
static MemoryContext MdCxt
Definition: md.c:88
void mdextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: md.c:414
#define SmgrIsTemp(smgr)
Definition: smgr.h:77
#define sprintf
Definition: port.h:217
void pfree(void *pointer)
Definition: mcxt.c:1057
static MdfdVec * mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
Definition: md.c:478
#define ERROR
Definition: elog.h:43
#define EXTENSION_RETURN_NULL
Definition: md.c:106
static void register_unlink_segment(RelFileNodeBackend rnode, ForkNumber forknum, BlockNumber segno)
Definition: md.c:1000
#define FATAL
Definition: elog.h:52
#define MAXPGPATH
void DropRelationFiles(RelFileNode *delrels, int ndelrels, bool isRedo)
Definition: md.c:1049
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:192
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
int FileSync(File file, uint32 wait_event_info)
Definition: fd.c:2154
static void mdunlinkfork(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
Definition: md.c:314
void mdunlink(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
Definition: md.c:277
int errcode_for_file_access(void)
Definition: elog.c:714
#define EXTENSION_CREATE_RECOVERY
Definition: md.c:110
void mdwriteback(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, BlockNumber nblocks)
Definition: md.c:583
MemoryContext TopMemoryContext
Definition: mcxt.c:44
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:146
ForkNumber
Definition: relpath.h:40
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:45
#define WARNING
Definition: elog.h:40
int mdunlinkfiletag(const FileTag *ftag, char *path)
Definition: md.c:1380
void mdopen(SMgrRelation reln)
Definition: md.c:521
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2056
#define InvalidBackendId
Definition: backendid.h:23
void * palloc0(Size size)
Definition: mcxt.c:981
int data_sync_elevel(int elevel)
Definition: fd.c:3630
static void _fdvec_resize(SMgrRelation reln, ForkNumber forknum, int nseg)
Definition: md.c:1081
bool RegisterSyncRequest(const FileTag *ftag, SyncRequestType type, bool retryOnError)
Definition: sync.c:554
Definition: md.c:82
bool mdfiletagmatches(const FileTag *ftag, const FileTag *candidate)
Definition: md.c:1399
static MdfdVec * _mdfd_openseg(SMgrRelation reln, ForkNumber forkno, BlockNumber segno, int oflags)
Definition: md.c:1142
void mdtruncate(SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
Definition: md.c:819
#define ereport(elevel,...)
Definition: elog.h:155
#define EXTENSION_CREATE
Definition: md.c:108
bool mdexists(SMgrRelation reln, ForkNumber forkNum)
Definition: md.c:161
RelFileNode node
Definition: relfilenode.h:74
void ForgetDatabaseSyncRequests(Oid dbid)
Definition: md.c:1031
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
void FileClose(File file)
Definition: fd.c:1853
int FilePrefetch(File file, off_t offset, int amount, uint32 wait_event_info)
Definition: fd.c:1949
void mdwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: md.c:697
#define Assert(condition)
Definition: c.h:800
void FileWriteback(File file, off_t offset, off_t nbytes, uint32 wait_event_info)
Definition: fd.c:1977
void TablespaceCreateDbspace(Oid spcNode, Oid dbNode, bool isRedo)
Definition: tablespace.c:115
static BlockNumber _mdnblocks(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:1312
BackendId backend
Definition: relfilenode.h:75
#define InvalidBlockNumber
Definition: block.h:33
#define MAX_FORKNUM
Definition: relpath.h:55
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1070
int mdsyncfiletag(const FileTag *ftag, char *path)
Definition: md.c:1333
void mdclose(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:532
bool mdprefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
Definition: md.c:555
void XLogDropRelation(RelFileNode rnode, ForkNumber forknum)
Definition: xlogutils.c:614
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
void * palloc(Size size)
Definition: mcxt.c:950
int errmsg(const char *fmt,...)
Definition: elog.c:902
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:797
#define elog(elevel,...)
Definition: elog.h:228
int i
void mdcreate(SMgrRelation reln, ForkNumber forkNum, bool isRedo)
Definition: md.c:178
#define relpath(rnode, forknum)
Definition: relpath.h:87
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69
int FileRead(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2000
int FileTruncate(File file, off_t offset, uint32 wait_event_info)
Definition: fd.c:2192
File mdfd_vfd
Definition: md.c:84
int File
Definition: fd.h:49
Definition: sync.h:50
#define FILE_POSSIBLY_DELETED(err)
Definition: fd.h:69
static void register_dirty_segment(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:974
static char * _mdfd_segpath(SMgrRelation reln, ForkNumber forknum, BlockNumber segno)
Definition: md.c:1119
bool zero_damaged_pages
Definition: bufmgr.c:124
void mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer)
Definition: md.c:632