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 static void
290 mdunlinkfork(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
291 {
292  char *path;
293  int ret;
294 
295  path = relpath(rnode, forkNum);
296 
297  /*
298  * Delete or truncate the first segment.
299  */
300  if (isRedo || forkNum != MAIN_FORKNUM || RelFileNodeBackendIsTemp(rnode))
301  {
302  /* First, forget any pending sync requests for the first segment */
303  if (!RelFileNodeBackendIsTemp(rnode))
304  register_forget_request(rnode, forkNum, 0 /* first seg */ );
305 
306  /* Next unlink the file */
307  ret = unlink(path);
308  if (ret < 0 && errno != ENOENT)
311  errmsg("could not remove file \"%s\": %m", path)));
312  }
313  else
314  {
315  /* truncate(2) would be easier here, but Windows hasn't got it */
316  int fd;
317 
318  fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
319  if (fd >= 0)
320  {
321  int save_errno;
322 
323  ret = ftruncate(fd, 0);
324  save_errno = errno;
325  CloseTransientFile(fd);
326  errno = save_errno;
327  }
328  else
329  ret = -1;
330  if (ret < 0 && errno != ENOENT)
333  errmsg("could not truncate file \"%s\": %m", path)));
334 
335  /* Register request to unlink first segment later */
336  register_unlink_segment(rnode, forkNum, 0 /* first seg */ );
337  }
338 
339  /*
340  * Delete any additional segments.
341  */
342  if (ret >= 0)
343  {
344  char *segpath = (char *) palloc(strlen(path) + 12);
345  BlockNumber segno;
346 
347  /*
348  * Note that because we loop until getting ENOENT, we will correctly
349  * remove all inactive segments as well as active ones.
350  */
351  for (segno = 1;; segno++)
352  {
353  /*
354  * Forget any pending sync requests for this segment before we try
355  * to unlink.
356  */
357  if (!RelFileNodeBackendIsTemp(rnode))
358  register_forget_request(rnode, forkNum, segno);
359 
360  sprintf(segpath, "%s.%u", path, segno);
361  if (unlink(segpath) < 0)
362  {
363  /* ENOENT is expected after the last segment... */
364  if (errno != ENOENT)
367  errmsg("could not remove file \"%s\": %m", segpath)));
368  break;
369  }
370  }
371  pfree(segpath);
372  }
373 
374  pfree(path);
375 }
376 
377 /*
378  * mdextend() -- Add a block to the specified relation.
379  *
380  * The semantics are nearly the same as mdwrite(): write at the
381  * specified position. However, this is to be used for the case of
382  * extending a relation (i.e., blocknum is at or beyond the current
383  * EOF). Note that we assume writing a block beyond current EOF
384  * causes intervening file space to become filled with zeroes.
385  */
386 void
387 mdextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
388  char *buffer, bool skipFsync)
389 {
390  off_t seekpos;
391  int nbytes;
392  MdfdVec *v;
393 
394  /* This assert is too expensive to have on normally ... */
395 #ifdef CHECK_WRITE_VS_EXTEND
396  Assert(blocknum >= mdnblocks(reln, forknum));
397 #endif
398 
399  /*
400  * If a relation manages to grow to 2^32-1 blocks, refuse to extend it any
401  * more --- we mustn't create a block whose number actually is
402  * InvalidBlockNumber.
403  */
404  if (blocknum == InvalidBlockNumber)
405  ereport(ERROR,
406  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
407  errmsg("cannot extend file \"%s\" beyond %u blocks",
408  relpath(reln->smgr_rnode, forknum),
410 
411  v = _mdfd_getseg(reln, forknum, blocknum, skipFsync, EXTENSION_CREATE);
412 
413  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
414 
415  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
416 
417  if ((nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_EXTEND)) != BLCKSZ)
418  {
419  if (nbytes < 0)
420  ereport(ERROR,
422  errmsg("could not extend file \"%s\": %m",
423  FilePathName(v->mdfd_vfd)),
424  errhint("Check free disk space.")));
425  /* short write: complain appropriately */
426  ereport(ERROR,
427  (errcode(ERRCODE_DISK_FULL),
428  errmsg("could not extend file \"%s\": wrote only %d of %d bytes at block %u",
430  nbytes, BLCKSZ, blocknum),
431  errhint("Check free disk space.")));
432  }
433 
434  if (!skipFsync && !SmgrIsTemp(reln))
435  register_dirty_segment(reln, forknum, v);
436 
437  Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
438 }
439 
440 /*
441  * mdopenfork() -- Open one fork of the specified relation.
442  *
443  * Note we only open the first segment, when there are multiple segments.
444  *
445  * If first segment is not present, either ereport or return NULL according
446  * to "behavior". We treat EXTENSION_CREATE the same as EXTENSION_FAIL;
447  * EXTENSION_CREATE means it's OK to extend an existing relation, not to
448  * invent one out of whole cloth.
449  */
450 static MdfdVec *
451 mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
452 {
453  MdfdVec *mdfd;
454  char *path;
455  File fd;
456 
457  /* No work if already open */
458  if (reln->md_num_open_segs[forknum] > 0)
459  return &reln->md_seg_fds[forknum][0];
460 
461  path = relpath(reln->smgr_rnode, forknum);
462 
463  fd = PathNameOpenFile(path, O_RDWR | PG_BINARY);
464 
465  if (fd < 0)
466  {
467  if ((behavior & EXTENSION_RETURN_NULL) &&
468  FILE_POSSIBLY_DELETED(errno))
469  {
470  pfree(path);
471  return NULL;
472  }
473  ereport(ERROR,
475  errmsg("could not open file \"%s\": %m", path)));
476  }
477 
478  pfree(path);
479 
480  _fdvec_resize(reln, forknum, 1);
481  mdfd = &reln->md_seg_fds[forknum][0];
482  mdfd->mdfd_vfd = fd;
483  mdfd->mdfd_segno = 0;
484 
485  Assert(_mdnblocks(reln, forknum, mdfd) <= ((BlockNumber) RELSEG_SIZE));
486 
487  return mdfd;
488 }
489 
490 /*
491  * mdopen() -- Initialize newly-opened relation.
492  */
493 void
495 {
496  /* mark it not open */
497  for (int forknum = 0; forknum <= MAX_FORKNUM; forknum++)
498  reln->md_num_open_segs[forknum] = 0;
499 }
500 
501 /*
502  * mdclose() -- Close the specified relation, if it isn't closed already.
503  */
504 void
506 {
507  int nopensegs = reln->md_num_open_segs[forknum];
508 
509  /* No work if already closed */
510  if (nopensegs == 0)
511  return;
512 
513  /* close segments starting from the end */
514  while (nopensegs > 0)
515  {
516  MdfdVec *v = &reln->md_seg_fds[forknum][nopensegs - 1];
517 
518  FileClose(v->mdfd_vfd);
519  _fdvec_resize(reln, forknum, nopensegs - 1);
520  nopensegs--;
521  }
522 }
523 
524 /*
525  * mdprefetch() -- Initiate asynchronous read of the specified block of a relation
526  */
527 bool
529 {
530 #ifdef USE_PREFETCH
531  off_t seekpos;
532  MdfdVec *v;
533 
534  v = _mdfd_getseg(reln, forknum, blocknum, false,
536  if (v == NULL)
537  return false;
538 
539  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
540 
541  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
542 
543  (void) FilePrefetch(v->mdfd_vfd, seekpos, BLCKSZ, WAIT_EVENT_DATA_FILE_PREFETCH);
544 #endif /* USE_PREFETCH */
545 
546  return true;
547 }
548 
549 /*
550  * mdwriteback() -- Tell the kernel to write pages back to storage.
551  *
552  * This accepts a range of blocks because flushing several pages at once is
553  * considerably more efficient than doing so individually.
554  */
555 void
557  BlockNumber blocknum, BlockNumber nblocks)
558 {
559  /*
560  * Issue flush requests in as few requests as possible; have to split at
561  * segment boundaries though, since those are actually separate files.
562  */
563  while (nblocks > 0)
564  {
565  BlockNumber nflush = nblocks;
566  off_t seekpos;
567  MdfdVec *v;
568  int segnum_start,
569  segnum_end;
570 
571  v = _mdfd_getseg(reln, forknum, blocknum, true /* not used */ ,
573 
574  /*
575  * We might be flushing buffers of already removed relations, that's
576  * ok, just ignore that case.
577  */
578  if (!v)
579  return;
580 
581  /* compute offset inside the current segment */
582  segnum_start = blocknum / RELSEG_SIZE;
583 
584  /* compute number of desired writes within the current segment */
585  segnum_end = (blocknum + nblocks - 1) / RELSEG_SIZE;
586  if (segnum_start != segnum_end)
587  nflush = RELSEG_SIZE - (blocknum % ((BlockNumber) RELSEG_SIZE));
588 
589  Assert(nflush >= 1);
590  Assert(nflush <= nblocks);
591 
592  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
593 
594  FileWriteback(v->mdfd_vfd, seekpos, (off_t) BLCKSZ * nflush, WAIT_EVENT_DATA_FILE_FLUSH);
595 
596  nblocks -= nflush;
597  blocknum += nflush;
598  }
599 }
600 
601 /*
602  * mdread() -- Read the specified block from a relation.
603  */
604 void
605 mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
606  char *buffer)
607 {
608  off_t seekpos;
609  int nbytes;
610  MdfdVec *v;
611 
612  TRACE_POSTGRESQL_SMGR_MD_READ_START(forknum, blocknum,
613  reln->smgr_rnode.node.spcNode,
614  reln->smgr_rnode.node.dbNode,
615  reln->smgr_rnode.node.relNode,
616  reln->smgr_rnode.backend);
617 
618  v = _mdfd_getseg(reln, forknum, blocknum, false,
620 
621  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
622 
623  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
624 
625  nbytes = FileRead(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_READ);
626 
627  TRACE_POSTGRESQL_SMGR_MD_READ_DONE(forknum, blocknum,
628  reln->smgr_rnode.node.spcNode,
629  reln->smgr_rnode.node.dbNode,
630  reln->smgr_rnode.node.relNode,
631  reln->smgr_rnode.backend,
632  nbytes,
633  BLCKSZ);
634 
635  if (nbytes != BLCKSZ)
636  {
637  if (nbytes < 0)
638  ereport(ERROR,
640  errmsg("could not read block %u in file \"%s\": %m",
641  blocknum, FilePathName(v->mdfd_vfd))));
642 
643  /*
644  * Short read: we are at or past EOF, or we read a partial block at
645  * EOF. Normally this is an error; upper levels should never try to
646  * read a nonexistent block. However, if zero_damaged_pages is ON or
647  * we are InRecovery, we should instead return zeroes without
648  * complaining. This allows, for example, the case of trying to
649  * update a block that was later truncated away.
650  */
652  MemSet(buffer, 0, BLCKSZ);
653  else
654  ereport(ERROR,
656  errmsg("could not read block %u in file \"%s\": read only %d of %d bytes",
657  blocknum, FilePathName(v->mdfd_vfd),
658  nbytes, BLCKSZ)));
659  }
660 }
661 
662 /*
663  * mdwrite() -- Write the supplied block at the appropriate location.
664  *
665  * This is to be used only for updating already-existing blocks of a
666  * relation (ie, those before the current EOF). To extend a relation,
667  * use mdextend().
668  */
669 void
670 mdwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
671  char *buffer, bool skipFsync)
672 {
673  off_t seekpos;
674  int nbytes;
675  MdfdVec *v;
676 
677  /* This assert is too expensive to have on normally ... */
678 #ifdef CHECK_WRITE_VS_EXTEND
679  Assert(blocknum < mdnblocks(reln, forknum));
680 #endif
681 
682  TRACE_POSTGRESQL_SMGR_MD_WRITE_START(forknum, blocknum,
683  reln->smgr_rnode.node.spcNode,
684  reln->smgr_rnode.node.dbNode,
685  reln->smgr_rnode.node.relNode,
686  reln->smgr_rnode.backend);
687 
688  v = _mdfd_getseg(reln, forknum, blocknum, skipFsync,
690 
691  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
692 
693  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
694 
695  nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_WRITE);
696 
697  TRACE_POSTGRESQL_SMGR_MD_WRITE_DONE(forknum, blocknum,
698  reln->smgr_rnode.node.spcNode,
699  reln->smgr_rnode.node.dbNode,
700  reln->smgr_rnode.node.relNode,
701  reln->smgr_rnode.backend,
702  nbytes,
703  BLCKSZ);
704 
705  if (nbytes != BLCKSZ)
706  {
707  if (nbytes < 0)
708  ereport(ERROR,
710  errmsg("could not write block %u in file \"%s\": %m",
711  blocknum, FilePathName(v->mdfd_vfd))));
712  /* short write: complain appropriately */
713  ereport(ERROR,
714  (errcode(ERRCODE_DISK_FULL),
715  errmsg("could not write block %u in file \"%s\": wrote only %d of %d bytes",
716  blocknum,
718  nbytes, BLCKSZ),
719  errhint("Check free disk space.")));
720  }
721 
722  if (!skipFsync && !SmgrIsTemp(reln))
723  register_dirty_segment(reln, forknum, v);
724 }
725 
726 /*
727  * mdnblocks() -- Get the number of blocks stored in a relation.
728  *
729  * Important side effect: all active segments of the relation are opened
730  * and added to the md_seg_fds array. If this routine has not been
731  * called, then only segments up to the last one actually touched
732  * are present in the array.
733  */
736 {
737  MdfdVec *v;
738  BlockNumber nblocks;
739  BlockNumber segno;
740 
741  mdopenfork(reln, forknum, EXTENSION_FAIL);
742 
743  /* mdopen has opened the first segment */
744  Assert(reln->md_num_open_segs[forknum] > 0);
745 
746  /*
747  * Start from the last open segments, to avoid redundant seeks. We have
748  * previously verified that these segments are exactly RELSEG_SIZE long,
749  * and it's useless to recheck that each time.
750  *
751  * NOTE: this assumption could only be wrong if another backend has
752  * truncated the relation. We rely on higher code levels to handle that
753  * scenario by closing and re-opening the md fd, which is handled via
754  * relcache flush. (Since the checkpointer doesn't participate in
755  * relcache flush, it could have segment entries for inactive segments;
756  * that's OK because the checkpointer never needs to compute relation
757  * size.)
758  */
759  segno = reln->md_num_open_segs[forknum] - 1;
760  v = &reln->md_seg_fds[forknum][segno];
761 
762  for (;;)
763  {
764  nblocks = _mdnblocks(reln, forknum, v);
765  if (nblocks > ((BlockNumber) RELSEG_SIZE))
766  elog(FATAL, "segment too big");
767  if (nblocks < ((BlockNumber) RELSEG_SIZE))
768  return (segno * ((BlockNumber) RELSEG_SIZE)) + nblocks;
769 
770  /*
771  * If segment is exactly RELSEG_SIZE, advance to next one.
772  */
773  segno++;
774 
775  /*
776  * We used to pass O_CREAT here, but that has the disadvantage that it
777  * might create a segment which has vanished through some operating
778  * system misadventure. In such a case, creating the segment here
779  * undermines _mdfd_getseg's attempts to notice and report an error
780  * upon access to a missing segment.
781  */
782  v = _mdfd_openseg(reln, forknum, segno, 0);
783  if (v == NULL)
784  return segno * ((BlockNumber) RELSEG_SIZE);
785  }
786 }
787 
788 /*
789  * mdtruncate() -- Truncate relation to specified number of blocks.
790  */
791 void
793 {
794  BlockNumber curnblk;
795  BlockNumber priorblocks;
796  int curopensegs;
797 
798  /*
799  * NOTE: mdnblocks makes sure we have opened all active segments, so that
800  * truncation loop will get them all!
801  */
802  curnblk = mdnblocks(reln, forknum);
803  if (nblocks > curnblk)
804  {
805  /* Bogus request ... but no complaint if InRecovery */
806  if (InRecovery)
807  return;
808  ereport(ERROR,
809  (errmsg("could not truncate file \"%s\" to %u blocks: it's only %u blocks now",
810  relpath(reln->smgr_rnode, forknum),
811  nblocks, curnblk)));
812  }
813  if (nblocks == curnblk)
814  return; /* no work */
815 
816  /*
817  * Truncate segments, starting at the last one. Starting at the end makes
818  * managing the memory for the fd array easier, should there be errors.
819  */
820  curopensegs = reln->md_num_open_segs[forknum];
821  while (curopensegs > 0)
822  {
823  MdfdVec *v;
824 
825  priorblocks = (curopensegs - 1) * RELSEG_SIZE;
826 
827  v = &reln->md_seg_fds[forknum][curopensegs - 1];
828 
829  if (priorblocks > nblocks)
830  {
831  /*
832  * This segment is no longer active. We truncate the file, but do
833  * not delete it, for reasons explained in the header comments.
834  */
836  ereport(ERROR,
838  errmsg("could not truncate file \"%s\": %m",
839  FilePathName(v->mdfd_vfd))));
840 
841  if (!SmgrIsTemp(reln))
842  register_dirty_segment(reln, forknum, v);
843 
844  /* we never drop the 1st segment */
845  Assert(v != &reln->md_seg_fds[forknum][0]);
846 
847  FileClose(v->mdfd_vfd);
848  _fdvec_resize(reln, forknum, curopensegs - 1);
849  }
850  else if (priorblocks + ((BlockNumber) RELSEG_SIZE) > nblocks)
851  {
852  /*
853  * This is the last segment we want to keep. Truncate the file to
854  * the right length. NOTE: if nblocks is exactly a multiple K of
855  * RELSEG_SIZE, we will truncate the K+1st segment to 0 length but
856  * keep it. This adheres to the invariant given in the header
857  * comments.
858  */
859  BlockNumber lastsegblocks = nblocks - priorblocks;
860 
861  if (FileTruncate(v->mdfd_vfd, (off_t) lastsegblocks * BLCKSZ, WAIT_EVENT_DATA_FILE_TRUNCATE) < 0)
862  ereport(ERROR,
864  errmsg("could not truncate file \"%s\" to %u blocks: %m",
866  nblocks)));
867  if (!SmgrIsTemp(reln))
868  register_dirty_segment(reln, forknum, v);
869  }
870  else
871  {
872  /*
873  * We still need this segment, so nothing to do for this and any
874  * earlier segment.
875  */
876  break;
877  }
878  curopensegs--;
879  }
880 }
881 
882 /*
883  * mdimmedsync() -- Immediately sync a relation to stable storage.
884  *
885  * Note that only writes already issued are synced; this routine knows
886  * nothing of dirty buffers that may exist inside the buffer manager. We
887  * sync active and inactive segments; smgrDoPendingSyncs() relies on this.
888  * Consider a relation skipping WAL. Suppose a checkpoint syncs blocks of
889  * some segment, then mdtruncate() renders that segment inactive. If we
890  * crash before the next checkpoint syncs the newly-inactive segment, that
891  * segment may survive recovery, reintroducing unwanted data into the table.
892  */
893 void
895 {
896  int segno;
897  int min_inactive_seg;
898 
899  /*
900  * NOTE: mdnblocks makes sure we have opened all active segments, so that
901  * fsync loop will get them all!
902  */
903  mdnblocks(reln, forknum);
904 
905  min_inactive_seg = segno = reln->md_num_open_segs[forknum];
906 
907  /*
908  * Temporarily open inactive segments, then close them after sync. There
909  * may be some inactive segments left opened after fsync() error, but that
910  * is harmless. We don't bother to clean them up and take a risk of
911  * further trouble. The next mdclose() will soon close them.
912  */
913  while (_mdfd_openseg(reln, forknum, segno, 0) != NULL)
914  segno++;
915 
916  while (segno > 0)
917  {
918  MdfdVec *v = &reln->md_seg_fds[forknum][segno - 1];
919 
923  errmsg("could not fsync file \"%s\": %m",
924  FilePathName(v->mdfd_vfd))));
925 
926  /* Close inactive segments immediately */
927  if (segno > min_inactive_seg)
928  {
929  FileClose(v->mdfd_vfd);
930  _fdvec_resize(reln, forknum, segno - 1);
931  }
932 
933  segno--;
934  }
935 }
936 
937 /*
938  * register_dirty_segment() -- Mark a relation segment as needing fsync
939  *
940  * If there is a local pending-ops table, just make an entry in it for
941  * ProcessSyncRequests to process later. Otherwise, try to pass off the
942  * fsync request to the checkpointer process. If that fails, just do the
943  * fsync locally before returning (we hope this will not happen often
944  * enough to be a performance problem).
945  */
946 static void
948 {
949  FileTag tag;
950 
951  INIT_MD_FILETAG(tag, reln->smgr_rnode.node, forknum, seg->mdfd_segno);
952 
953  /* Temp relations should never be fsync'd */
954  Assert(!SmgrIsTemp(reln));
955 
956  if (!RegisterSyncRequest(&tag, SYNC_REQUEST, false /* retryOnError */ ))
957  {
958  ereport(DEBUG1,
959  (errmsg("could not forward fsync request because request queue is full")));
960 
964  errmsg("could not fsync file \"%s\": %m",
965  FilePathName(seg->mdfd_vfd))));
966  }
967 }
968 
969 /*
970  * register_unlink_segment() -- Schedule a file to be deleted after next checkpoint
971  */
972 static void
974  BlockNumber segno)
975 {
976  FileTag tag;
977 
978  INIT_MD_FILETAG(tag, rnode.node, forknum, segno);
979 
980  /* Should never be used with temp relations */
982 
983  RegisterSyncRequest(&tag, SYNC_UNLINK_REQUEST, true /* retryOnError */ );
984 }
985 
986 /*
987  * register_forget_request() -- forget any fsyncs for a relation fork's segment
988  */
989 static void
991  BlockNumber segno)
992 {
993  FileTag tag;
994 
995  INIT_MD_FILETAG(tag, rnode.node, forknum, segno);
996 
997  RegisterSyncRequest(&tag, SYNC_FORGET_REQUEST, true /* retryOnError */ );
998 }
999 
1000 /*
1001  * ForgetDatabaseSyncRequests -- forget any fsyncs and unlinks for a DB
1002  */
1003 void
1005 {
1006  FileTag tag;
1007  RelFileNode rnode;
1008 
1009  rnode.dbNode = dbid;
1010  rnode.spcNode = 0;
1011  rnode.relNode = 0;
1012 
1014 
1015  RegisterSyncRequest(&tag, SYNC_FILTER_REQUEST, true /* retryOnError */ );
1016 }
1017 
1018 /*
1019  * DropRelationFiles -- drop files of all given relations
1020  */
1021 void
1022 DropRelationFiles(RelFileNode *delrels, int ndelrels, bool isRedo)
1023 {
1024  SMgrRelation *srels;
1025  int i;
1026 
1027  srels = palloc(sizeof(SMgrRelation) * ndelrels);
1028  for (i = 0; i < ndelrels; i++)
1029  {
1030  SMgrRelation srel = smgropen(delrels[i], InvalidBackendId);
1031 
1032  if (isRedo)
1033  {
1034  ForkNumber fork;
1035 
1036  for (fork = 0; fork <= MAX_FORKNUM; fork++)
1037  XLogDropRelation(delrels[i], fork);
1038  }
1039  srels[i] = srel;
1040  }
1041 
1042  smgrdounlinkall(srels, ndelrels, isRedo);
1043 
1044  for (i = 0; i < ndelrels; i++)
1045  smgrclose(srels[i]);
1046  pfree(srels);
1047 }
1048 
1049 
1050 /*
1051  * _fdvec_resize() -- Resize the fork's open segments array
1052  */
1053 static void
1055  ForkNumber forknum,
1056  int nseg)
1057 {
1058  if (nseg == 0)
1059  {
1060  if (reln->md_num_open_segs[forknum] > 0)
1061  {
1062  pfree(reln->md_seg_fds[forknum]);
1063  reln->md_seg_fds[forknum] = NULL;
1064  }
1065  }
1066  else if (reln->md_num_open_segs[forknum] == 0)
1067  {
1068  reln->md_seg_fds[forknum] =
1069  MemoryContextAlloc(MdCxt, sizeof(MdfdVec) * nseg);
1070  }
1071  else
1072  {
1073  /*
1074  * It doesn't seem worthwhile complicating the code to amortize
1075  * repalloc() calls. Those are far faster than PathNameOpenFile() or
1076  * FileClose(), and the memory context internally will sometimes avoid
1077  * doing an actual reallocation.
1078  */
1079  reln->md_seg_fds[forknum] =
1080  repalloc(reln->md_seg_fds[forknum],
1081  sizeof(MdfdVec) * nseg);
1082  }
1083 
1084  reln->md_num_open_segs[forknum] = nseg;
1085 }
1086 
1087 /*
1088  * Return the filename for the specified segment of the relation. The
1089  * returned string is palloc'd.
1090  */
1091 static char *
1093 {
1094  char *path,
1095  *fullpath;
1096 
1097  path = relpath(reln->smgr_rnode, forknum);
1098 
1099  if (segno > 0)
1100  {
1101  fullpath = psprintf("%s.%u", path, segno);
1102  pfree(path);
1103  }
1104  else
1105  fullpath = path;
1106 
1107  return fullpath;
1108 }
1109 
1110 /*
1111  * Open the specified segment of the relation,
1112  * and make a MdfdVec object for it. Returns NULL on failure.
1113  */
1114 static MdfdVec *
1116  int oflags)
1117 {
1118  MdfdVec *v;
1119  File fd;
1120  char *fullpath;
1121 
1122  fullpath = _mdfd_segpath(reln, forknum, segno);
1123 
1124  /* open the file */
1125  fd = PathNameOpenFile(fullpath, O_RDWR | PG_BINARY | oflags);
1126 
1127  pfree(fullpath);
1128 
1129  if (fd < 0)
1130  return NULL;
1131 
1132  /*
1133  * Segments are always opened in order from lowest to highest, so we must
1134  * be adding a new one at the end.
1135  */
1136  Assert(segno == reln->md_num_open_segs[forknum]);
1137 
1138  _fdvec_resize(reln, forknum, segno + 1);
1139 
1140  /* fill the entry */
1141  v = &reln->md_seg_fds[forknum][segno];
1142  v->mdfd_vfd = fd;
1143  v->mdfd_segno = segno;
1144 
1145  Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
1146 
1147  /* all done */
1148  return v;
1149 }
1150 
1151 /*
1152  * _mdfd_getseg() -- Find the segment of the relation holding the
1153  * specified block.
1154  *
1155  * If the segment doesn't exist, we ereport, return NULL, or create the
1156  * segment, according to "behavior". Note: skipFsync is only used in the
1157  * EXTENSION_CREATE case.
1158  */
1159 static MdfdVec *
1161  bool skipFsync, int behavior)
1162 {
1163  MdfdVec *v;
1164  BlockNumber targetseg;
1165  BlockNumber nextsegno;
1166 
1167  /* some way to handle non-existent segments needs to be specified */
1168  Assert(behavior &
1170 
1171  targetseg = blkno / ((BlockNumber) RELSEG_SIZE);
1172 
1173  /* if an existing and opened segment, we're done */
1174  if (targetseg < reln->md_num_open_segs[forknum])
1175  {
1176  v = &reln->md_seg_fds[forknum][targetseg];
1177  return v;
1178  }
1179 
1180  /*
1181  * The target segment is not yet open. Iterate over all the segments
1182  * between the last opened and the target segment. This way missing
1183  * segments either raise an error, or get created (according to
1184  * 'behavior'). Start with either the last opened, or the first segment if
1185  * none was opened before.
1186  */
1187  if (reln->md_num_open_segs[forknum] > 0)
1188  v = &reln->md_seg_fds[forknum][reln->md_num_open_segs[forknum] - 1];
1189  else
1190  {
1191  v = mdopenfork(reln, forknum, behavior);
1192  if (!v)
1193  return NULL; /* if behavior & EXTENSION_RETURN_NULL */
1194  }
1195 
1196  for (nextsegno = reln->md_num_open_segs[forknum];
1197  nextsegno <= targetseg; nextsegno++)
1198  {
1199  BlockNumber nblocks = _mdnblocks(reln, forknum, v);
1200  int flags = 0;
1201 
1202  Assert(nextsegno == v->mdfd_segno + 1);
1203 
1204  if (nblocks > ((BlockNumber) RELSEG_SIZE))
1205  elog(FATAL, "segment too big");
1206 
1207  if ((behavior & EXTENSION_CREATE) ||
1208  (InRecovery && (behavior & EXTENSION_CREATE_RECOVERY)))
1209  {
1210  /*
1211  * Normally we will create new segments only if authorized by the
1212  * caller (i.e., we are doing mdextend()). But when doing WAL
1213  * recovery, create segments anyway; this allows cases such as
1214  * replaying WAL data that has a write into a high-numbered
1215  * segment of a relation that was later deleted. We want to go
1216  * ahead and create the segments so we can finish out the replay.
1217  *
1218  * We have to maintain the invariant that segments before the last
1219  * active segment are of size RELSEG_SIZE; therefore, if
1220  * extending, pad them out with zeroes if needed. (This only
1221  * matters if in recovery, or if the caller is extending the
1222  * relation discontiguously, but that can happen in hash indexes.)
1223  */
1224  if (nblocks < ((BlockNumber) RELSEG_SIZE))
1225  {
1226  char *zerobuf = palloc0(BLCKSZ);
1227 
1228  mdextend(reln, forknum,
1229  nextsegno * ((BlockNumber) RELSEG_SIZE) - 1,
1230  zerobuf, skipFsync);
1231  pfree(zerobuf);
1232  }
1233  flags = O_CREAT;
1234  }
1235  else if (!(behavior & EXTENSION_DONT_CHECK_SIZE) &&
1236  nblocks < ((BlockNumber) RELSEG_SIZE))
1237  {
1238  /*
1239  * When not extending (or explicitly including truncated
1240  * segments), only open the next segment if the current one is
1241  * exactly RELSEG_SIZE. If not (this branch), either return NULL
1242  * or fail.
1243  */
1244  if (behavior & EXTENSION_RETURN_NULL)
1245  {
1246  /*
1247  * Some callers discern between reasons for _mdfd_getseg()
1248  * returning NULL based on errno. As there's no failing
1249  * syscall involved in this case, explicitly set errno to
1250  * ENOENT, as that seems the closest interpretation.
1251  */
1252  errno = ENOENT;
1253  return NULL;
1254  }
1255 
1256  ereport(ERROR,
1258  errmsg("could not open file \"%s\" (target block %u): previous segment is only %u blocks",
1259  _mdfd_segpath(reln, forknum, nextsegno),
1260  blkno, nblocks)));
1261  }
1262 
1263  v = _mdfd_openseg(reln, forknum, nextsegno, flags);
1264 
1265  if (v == NULL)
1266  {
1267  if ((behavior & EXTENSION_RETURN_NULL) &&
1268  FILE_POSSIBLY_DELETED(errno))
1269  return NULL;
1270  ereport(ERROR,
1272  errmsg("could not open file \"%s\" (target block %u): %m",
1273  _mdfd_segpath(reln, forknum, nextsegno),
1274  blkno)));
1275  }
1276  }
1277 
1278  return v;
1279 }
1280 
1281 /*
1282  * Get number of blocks present in a single disk file
1283  */
1284 static BlockNumber
1286 {
1287  off_t len;
1288 
1289  len = FileSize(seg->mdfd_vfd);
1290  if (len < 0)
1291  ereport(ERROR,
1293  errmsg("could not seek to end of file \"%s\": %m",
1294  FilePathName(seg->mdfd_vfd))));
1295  /* note that this calculation will ignore any partial block at EOF */
1296  return (BlockNumber) (len / BLCKSZ);
1297 }
1298 
1299 /*
1300  * Sync a file to disk, given a file tag. Write the path into an output
1301  * buffer so the caller can use it in error messages.
1302  *
1303  * Return 0 on success, -1 on failure, with errno set.
1304  */
1305 int
1306 mdsyncfiletag(const FileTag *ftag, char *path)
1307 {
1309  File file;
1310  bool need_to_close;
1311  int result,
1312  save_errno;
1313 
1314  /* See if we already have the file open, or need to open it. */
1315  if (ftag->segno < reln->md_num_open_segs[ftag->forknum])
1316  {
1317  file = reln->md_seg_fds[ftag->forknum][ftag->segno].mdfd_vfd;
1318  strlcpy(path, FilePathName(file), MAXPGPATH);
1319  need_to_close = false;
1320  }
1321  else
1322  {
1323  char *p;
1324 
1325  p = _mdfd_segpath(reln, ftag->forknum, ftag->segno);
1326  strlcpy(path, p, MAXPGPATH);
1327  pfree(p);
1328 
1329  file = PathNameOpenFile(path, O_RDWR | PG_BINARY);
1330  if (file < 0)
1331  return -1;
1332  need_to_close = true;
1333  }
1334 
1335  /* Sync the file. */
1336  result = FileSync(file, WAIT_EVENT_DATA_FILE_SYNC);
1337  save_errno = errno;
1338 
1339  if (need_to_close)
1340  FileClose(file);
1341 
1342  errno = save_errno;
1343  return result;
1344 }
1345 
1346 /*
1347  * Unlink a file, given a file tag. Write the path into an output
1348  * buffer so the caller can use it in error messages.
1349  *
1350  * Return 0 on success, -1 on failure, with errno set.
1351  */
1352 int
1353 mdunlinkfiletag(const FileTag *ftag, char *path)
1354 {
1355  char *p;
1356 
1357  /* Compute the path. */
1358  p = relpathperm(ftag->rnode, MAIN_FORKNUM);
1359  strlcpy(path, p, MAXPGPATH);
1360  pfree(p);
1361 
1362  /* Try to unlink the file. */
1363  return unlink(path);
1364 }
1365 
1366 /*
1367  * Check if a given candidate request matches a given tag, when processing
1368  * a SYNC_FILTER_REQUEST request. This will be called for all pending
1369  * requests to find out whether to forget them.
1370  */
1371 bool
1372 mdfiletagmatches(const FileTag *ftag, const FileTag *candidate)
1373 {
1374  /*
1375  * For now we only use filter requests as a way to drop all scheduled
1376  * callbacks relating to a given database, when dropping the database.
1377  * We'll return true for all candidates that have the same database OID as
1378  * the ftag from the SYNC_FILTER_REQUEST request, so they're forgotten.
1379  */
1380  return ftag->rnode.dbNode == candidate->rnode.dbNode;
1381 }
uint32 segno
Definition: sync.h:50
#define relpathperm(rnode, forknum)
Definition: relpath.h:83
void mdimmedsync(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:894
#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:1160
int errhint(const char *fmt,...)
Definition: elog.c:1071
#define EXTENSION_DONT_CHECK_SIZE
Definition: md.c:118
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1437
#define RelFileNodeBackendIsTemp(rnode)
Definition: relfilenode.h:78
BlockNumber mdnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:735
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:2148
int16 forknum
Definition: sync.h:48
int errcode(int sqlerrcode)
Definition: elog.c:610
void smgrdounlinkall(SMgrRelation *rels, int nrels, bool isRedo)
Definition: smgr.c:385
#define MemSet(start, val, len)
Definition: c.h:949
RelFileNode rnode
Definition: sync.h:49
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:2200
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1211
#define EXTENSION_FAIL
Definition: md.c:104
static void register_forget_request(RelFileNodeBackend rnode, ForkNumber forknum, BlockNumber segno)
Definition: md.c:990
static MemoryContext MdCxt
Definition: md.c:88
void mdextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: md.c:387
#define SmgrIsTemp(smgr)
Definition: smgr.h:77
#define sprintf
Definition: port.h:195
void pfree(void *pointer)
Definition: mcxt.c:1057
static MdfdVec * mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
Definition: md.c:451
#define ERROR
Definition: elog.h:43
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2372
#define EXTENSION_RETURN_NULL
Definition: md.c:106
static void register_unlink_segment(RelFileNodeBackend rnode, ForkNumber forknum, BlockNumber segno)
Definition: md.c:973
#define FATAL
Definition: elog.h:52
#define MAXPGPATH
void DropRelationFiles(RelFileNode *delrels, int ndelrels, bool isRedo)
Definition: md.c:1022
#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:2127
static void mdunlinkfork(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
Definition: md.c:290
void mdunlink(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
Definition: md.c:277
int errcode_for_file_access(void)
Definition: elog.c:633
#define EXTENSION_CREATE_RECOVERY
Definition: md.c:110
void mdwriteback(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, BlockNumber nblocks)
Definition: md.c:556
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
int CloseTransientFile(int fd)
Definition: fd.c:2549
#define WARNING
Definition: elog.h:40
int mdunlinkfiletag(const FileTag *ftag, char *path)
Definition: md.c:1353
void mdopen(SMgrRelation reln)
Definition: md.c:494
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2029
#define InvalidBackendId
Definition: backendid.h:23
void * palloc0(Size size)
Definition: mcxt.c:981
int data_sync_elevel(int elevel)
Definition: fd.c:3603
static void _fdvec_resize(SMgrRelation reln, ForkNumber forknum, int nseg)
Definition: md.c:1054
bool RegisterSyncRequest(const FileTag *ftag, SyncRequestType type, bool retryOnError)
Definition: sync.c:532
Definition: md.c:82
bool mdfiletagmatches(const FileTag *ftag, const FileTag *candidate)
Definition: md.c:1372
static MdfdVec * _mdfd_openseg(SMgrRelation reln, ForkNumber forkno, BlockNumber segno, int oflags)
Definition: md.c:1115
void mdtruncate(SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
Definition: md.c:792
#define ereport(elevel,...)
Definition: elog.h:144
#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:1004
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
void FileClose(File file)
Definition: fd.c:1826
int FilePrefetch(File file, off_t offset, int amount, uint32 wait_event_info)
Definition: fd.c:1922
void mdwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: md.c:670
#define Assert(condition)
Definition: c.h:745
void FileWriteback(File file, off_t offset, off_t nbytes, uint32 wait_event_info)
Definition: fd.c:1950
void TablespaceCreateDbspace(Oid spcNode, Oid dbNode, bool isRedo)
Definition: tablespace.c:115
static BlockNumber _mdnblocks(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:1285
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:1306
void mdclose(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:505
bool mdprefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
Definition: md.c:528
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:824
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:797
#define elog(elevel,...)
Definition: elog.h:214
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:1973
int FileTruncate(File file, off_t offset, uint32 wait_event_info)
Definition: fd.c:2165
File mdfd_vfd
Definition: md.c:84
int File
Definition: fd.h:49
Definition: sync.h:45
#define FILE_POSSIBLY_DELETED(err)
Definition: fd.h:69
static void register_dirty_segment(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:947
static char * _mdfd_segpath(SMgrRelation reln, ForkNumber forknum, BlockNumber segno)
Definition: md.c:1092
bool zero_damaged_pages
Definition: bufmgr.c:124
void mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer)
Definition: md.c:605
#define ftruncate(a, b)
Definition: win32_port.h:59