1 /*-------------------------------------------------------------------------
4 * public interface routines to storage manager switch.
6 * All file system operations in POSTGRES dispatch through these
9 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
10 * Portions Copyright (c) 1994, Regents of the University of California
16 *-------------------------------------------------------------------------
20 #include "access/xact.h"
21 #include "access/xlogutils.h"
22 #include "commands/tablespace.h"
23 #include "storage/bufmgr.h"
24 #include "storage/ipc.h"
25 #include "storage/smgr.h"
26 #include "utils/hsearch.h"
27 #include "utils/memutils.h"
31 * This struct of function pointers defines the API between smgr.c and
32 * any individual storage manager module. Note that smgr subfunctions are
33 * generally expected to report problems via elog(ERROR). An exception is
34 * that smgr_unlink should use elog(WARNING), rather than erroring out,
35 * because we normally unlink relations during post-commit/abort cleanup,
36 * and so it's too late to raise an error. Also, various conditions that
37 * would normally be errors should be allowed during bootstrap and/or WAL
38 * recovery --- see comments in md.c for details.
42 void (*smgr_init
) (void); /* may be NULL */
43 void (*smgr_shutdown
) (void); /* may be NULL */
44 void (*smgr_close
) (SMgrRelation reln
, ForkNumber forknum
);
45 void (*smgr_create
) (SMgrRelation reln
, ForkNumber forknum
,
47 bool (*smgr_exists
) (SMgrRelation reln
, ForkNumber forknum
);
48 void (*smgr_unlink
) (RelFileNode rnode
, ForkNumber forknum
,
50 void (*smgr_extend
) (SMgrRelation reln
, ForkNumber forknum
,
51 BlockNumber blocknum
, char *buffer
, bool isTemp
);
52 void (*smgr_read
) (SMgrRelation reln
, ForkNumber forknum
,
53 BlockNumber blocknum
, char *buffer
);
54 void (*smgr_write
) (SMgrRelation reln
, ForkNumber forknum
,
55 BlockNumber blocknum
, char *buffer
, bool isTemp
);
56 BlockNumber (*smgr_nblocks
) (SMgrRelation reln
, ForkNumber forknum
);
57 void (*smgr_truncate
) (SMgrRelation reln
, ForkNumber forknum
,
58 BlockNumber nblocks
, bool isTemp
);
59 void (*smgr_immedsync
) (SMgrRelation reln
, ForkNumber forknum
);
60 void (*smgr_commit
) (void); /* may be NULL */
61 void (*smgr_abort
) (void); /* may be NULL */
62 void (*smgr_pre_ckpt
) (void); /* may be NULL */
63 void (*smgr_sync
) (void); /* may be NULL */
64 void (*smgr_post_ckpt
) (void); /* may be NULL */
68 static const f_smgr smgrsw
[] = {
70 {mdinit
, NULL
, mdclose
, mdcreate
, mdexists
, mdunlink
, mdextend
,
71 mdread
, mdwrite
, mdnblocks
, mdtruncate
, mdimmedsync
,
72 NULL
, NULL
, mdpreckpt
, mdsync
, mdpostckpt
76 static const int NSmgr
= lengthof(smgrsw
);
80 * Each backend has a hashtable that stores all extant SMgrRelation objects.
82 static HTAB
*SMgrRelationHash
= NULL
;
85 * We keep a list of all relations (represented as RelFileNode values)
86 * that have been created or deleted in the current transaction. When
87 * a relation is created, we create the physical file immediately, but
88 * remember it so that we can delete the file again if the current
89 * transaction is aborted. Conversely, a deletion request is NOT
90 * executed immediately, but is just entered in the list. When and if
91 * the transaction commits, we can delete the physical file.
93 * To handle subtransactions, every entry is marked with its transaction
94 * nesting level. At subtransaction commit, we reassign the subtransaction's
95 * entries to the parent nesting level. At subtransaction abort, we can
96 * immediately execute the abort-time actions for all entries of the current
99 * NOTE: the list is kept in TopMemoryContext to be sure it won't disappear
100 * unbetimes. It'd probably be OK to keep it in TopTransactionContext,
101 * but I'm being paranoid.
104 typedef struct PendingRelDelete
106 RelFileNode relnode
; /* relation that may need to be deleted */
107 ForkNumber forknum
; /* fork number that may need to be deleted */
108 int which
; /* which storage manager? */
109 bool isTemp
; /* is it a temporary relation? */
110 bool atCommit
; /* T=delete at commit; F=delete at abort */
111 int nestLevel
; /* xact nesting level of request */
112 struct PendingRelDelete
*next
; /* linked-list link */
115 static PendingRelDelete
*pendingDeletes
= NULL
; /* head of linked list */
119 * Declarations for smgr-related XLOG records
121 * Note: we log file creation and truncation here, but logging of deletion
122 * actions is handled by xact.c, because it is part of transaction commit.
125 /* XLOG gives us high 4 bits */
126 #define XLOG_SMGR_CREATE 0x10
127 #define XLOG_SMGR_TRUNCATE 0x20
129 typedef struct xl_smgr_create
135 typedef struct xl_smgr_truncate
143 /* local function prototypes */
144 static void smgrshutdown(int code
, Datum arg
);
145 static void smgr_internal_unlink(RelFileNode rnode
, ForkNumber forknum
,
146 int which
, bool isTemp
, bool isRedo
);
150 * smgrinit(), smgrshutdown() -- Initialize or shut down storage
153 * Note: smgrinit is called during backend startup (normal or standalone
154 * case), *not* during postmaster start. Therefore, any resources created
155 * here or destroyed in smgrshutdown are backend-local.
162 for (i
= 0; i
< NSmgr
; i
++)
164 if (smgrsw
[i
].smgr_init
)
165 (*(smgrsw
[i
].smgr_init
)) ();
168 /* register the shutdown proc */
169 on_proc_exit(smgrshutdown
, 0);
173 * on_proc_exit hook for smgr cleanup during backend shutdown
176 smgrshutdown(int code
, Datum arg
)
180 for (i
= 0; i
< NSmgr
; i
++)
182 if (smgrsw
[i
].smgr_shutdown
)
183 (*(smgrsw
[i
].smgr_shutdown
)) ();
188 * smgropen() -- Return an SMgrRelation object, creating it if need be.
190 * This does not attempt to actually open the object.
193 smgropen(RelFileNode rnode
)
198 if (SMgrRelationHash
== NULL
)
200 /* First time through: initialize the hash table */
203 MemSet(&ctl
, 0, sizeof(ctl
));
204 ctl
.keysize
= sizeof(RelFileNode
);
205 ctl
.entrysize
= sizeof(SMgrRelationData
);
207 SMgrRelationHash
= hash_create("smgr relation table", 400,
208 &ctl
, HASH_ELEM
| HASH_FUNCTION
);
211 /* Look up or create an entry */
212 reln
= (SMgrRelation
) hash_search(SMgrRelationHash
,
216 /* Initialize it if not present before */
221 /* hash_search already filled in the lookup key */
222 reln
->smgr_owner
= NULL
;
223 reln
->smgr_which
= 0; /* we only have md.c at present */
225 /* mark it not open */
226 for(forknum
= 0; forknum
<= MAX_FORKNUM
; forknum
++)
227 reln
->md_fd
[forknum
] = NULL
;
234 * smgrsetowner() -- Establish a long-lived reference to an SMgrRelation object
236 * There can be only one owner at a time; this is sufficient since currently
237 * the only such owners exist in the relcache.
240 smgrsetowner(SMgrRelation
*owner
, SMgrRelation reln
)
243 * First, unhook any old owner. (Normally there shouldn't be any, but it
244 * seems possible that this can happen during swap_relation_files()
245 * depending on the order of processing. It's ok to close the old
246 * relcache entry early in that case.)
248 if (reln
->smgr_owner
)
249 *(reln
->smgr_owner
) = NULL
;
251 /* Now establish the ownership relationship. */
252 reln
->smgr_owner
= owner
;
257 * smgrexists() -- Does the underlying file for a fork exist?
260 smgrexists(SMgrRelation reln
, ForkNumber forknum
)
262 return (*(smgrsw
[reln
->smgr_which
].smgr_exists
)) (reln
, forknum
);
266 * smgrclose() -- Close and delete an SMgrRelation object.
269 smgrclose(SMgrRelation reln
)
274 for (forknum
= 0; forknum
<= MAX_FORKNUM
; forknum
++)
275 (*(smgrsw
[reln
->smgr_which
].smgr_close
)) (reln
, forknum
);
277 owner
= reln
->smgr_owner
;
279 if (hash_search(SMgrRelationHash
,
280 (void *) &(reln
->smgr_rnode
),
281 HASH_REMOVE
, NULL
) == NULL
)
282 elog(ERROR
, "SMgrRelation hashtable corrupted");
285 * Unhook the owner pointer, if any. We do this last since in the remote
286 * possibility of failure above, the SMgrRelation object will still exist.
293 * smgrcloseall() -- Close all existing SMgrRelation objects.
298 HASH_SEQ_STATUS status
;
301 /* Nothing to do if hashtable not set up */
302 if (SMgrRelationHash
== NULL
)
305 hash_seq_init(&status
, SMgrRelationHash
);
307 while ((reln
= (SMgrRelation
) hash_seq_search(&status
)) != NULL
)
312 * smgrclosenode() -- Close SMgrRelation object for given RelFileNode,
315 * This has the same effects as smgrclose(smgropen(rnode)), but it avoids
316 * uselessly creating a hashtable entry only to drop it again when no
317 * such entry exists already.
320 smgrclosenode(RelFileNode rnode
)
324 /* Nothing to do if hashtable not set up */
325 if (SMgrRelationHash
== NULL
)
328 reln
= (SMgrRelation
) hash_search(SMgrRelationHash
,
336 * smgrcreate() -- Create a new relation.
338 * Given an already-created (but presumably unused) SMgrRelation,
339 * cause the underlying disk file or other storage for the fork
342 * If isRedo is true, it is okay for the underlying file to exist
343 * already because we are in a WAL replay sequence. In this case
344 * we should make no PendingRelDelete entry; the WAL sequence will
345 * tell whether to drop the file.
348 smgrcreate(SMgrRelation reln
, ForkNumber forknum
, bool isTemp
, bool isRedo
)
352 xl_smgr_create xlrec
;
353 PendingRelDelete
*pending
;
356 * Exit quickly in WAL replay mode if we've already opened the file.
357 * If it's open, it surely must exist.
359 if (isRedo
&& reln
->md_fd
[forknum
] != NULL
)
363 * We may be using the target table space for the first time in this
364 * database, so create a per-database subdirectory if needed.
366 * XXX this is a fairly ugly violation of module layering, but this seems
367 * to be the best place to put the check. Maybe TablespaceCreateDbspace
368 * should be here and not in commands/tablespace.c? But that would imply
369 * importing a lot of stuff that smgr.c oughtn't know, either.
371 TablespaceCreateDbspace(reln
->smgr_rnode
.spcNode
,
372 reln
->smgr_rnode
.dbNode
,
375 (*(smgrsw
[reln
->smgr_which
].smgr_create
)) (reln
, forknum
, isRedo
);
381 * Make an XLOG entry showing the file creation. If we abort, the file
382 * will be dropped at abort time.
384 xlrec
.rnode
= reln
->smgr_rnode
;
385 xlrec
.forknum
= forknum
;
387 rdata
.data
= (char *) &xlrec
;
388 rdata
.len
= sizeof(xlrec
);
389 rdata
.buffer
= InvalidBuffer
;
392 lsn
= XLogInsert(RM_SMGR_ID
, XLOG_SMGR_CREATE
, &rdata
);
394 /* Add the relation to the list of stuff to delete at abort */
395 pending
= (PendingRelDelete
*)
396 MemoryContextAlloc(TopMemoryContext
, sizeof(PendingRelDelete
));
397 pending
->relnode
= reln
->smgr_rnode
;
398 pending
->forknum
= forknum
;
399 pending
->which
= reln
->smgr_which
;
400 pending
->isTemp
= isTemp
;
401 pending
->atCommit
= false; /* delete if abort */
402 pending
->nestLevel
= GetCurrentTransactionNestLevel();
403 pending
->next
= pendingDeletes
;
404 pendingDeletes
= pending
;
408 * smgrscheduleunlink() -- Schedule unlinking a relation at xact commit.
410 * The fork is marked to be removed from the store if we successfully
411 * commit the current transaction.
414 smgrscheduleunlink(SMgrRelation reln
, ForkNumber forknum
, bool isTemp
)
416 PendingRelDelete
*pending
;
418 /* Add the relation to the list of stuff to delete at commit */
419 pending
= (PendingRelDelete
*)
420 MemoryContextAlloc(TopMemoryContext
, sizeof(PendingRelDelete
));
421 pending
->relnode
= reln
->smgr_rnode
;
422 pending
->forknum
= forknum
;
423 pending
->which
= reln
->smgr_which
;
424 pending
->isTemp
= isTemp
;
425 pending
->atCommit
= true; /* delete if commit */
426 pending
->nestLevel
= GetCurrentTransactionNestLevel();
427 pending
->next
= pendingDeletes
;
428 pendingDeletes
= pending
;
431 * NOTE: if the relation was created in this transaction, it will now be
432 * present in the pending-delete list twice, once with atCommit true and
433 * once with atCommit false. Hence, it will be physically deleted at end
434 * of xact in either case (and the other entry will be ignored by
435 * smgrDoPendingDeletes, so no error will occur). We could instead remove
436 * the existing list entry and delete the physical file immediately, but
437 * for now I'll keep the logic simple.
442 * smgrdounlink() -- Immediately unlink a relation.
444 * The specified fork of the relation is removed from the store. This
445 * should not be used during transactional operations, since it can't be
448 * If isRedo is true, it is okay for the underlying file to be gone
452 smgrdounlink(SMgrRelation reln
, ForkNumber forknum
, bool isTemp
, bool isRedo
)
454 RelFileNode rnode
= reln
->smgr_rnode
;
455 int which
= reln
->smgr_which
;
458 (*(smgrsw
[which
].smgr_close
)) (reln
, forknum
);
460 smgr_internal_unlink(rnode
, forknum
, which
, isTemp
, isRedo
);
464 * Shared subroutine that actually does the unlink ...
467 smgr_internal_unlink(RelFileNode rnode
, ForkNumber forknum
,
468 int which
, bool isTemp
, bool isRedo
)
471 * Get rid of any remaining buffers for the relation. bufmgr will just
472 * drop them without bothering to write the contents.
474 DropRelFileNodeBuffers(rnode
, forknum
, isTemp
, 0);
477 * It'd be nice to tell the stats collector to forget it immediately, too.
478 * But we can't because we don't know the OID (and in cases involving
479 * relfilenode swaps, it's not always clear which table OID to forget,
484 * And delete the physical files.
486 * Note: smgr_unlink must treat deletion failure as a WARNING, not an
487 * ERROR, because we've already decided to commit or abort the current
490 (*(smgrsw
[which
].smgr_unlink
)) (rnode
, forknum
, isRedo
);
494 * smgrextend() -- Add a new block to a file.
496 * The semantics are nearly the same as smgrwrite(): write at the
497 * specified position. However, this is to be used for the case of
498 * extending a relation (i.e., blocknum is at or beyond the current
499 * EOF). Note that we assume writing a block beyond current EOF
500 * causes intervening file space to become filled with zeroes.
503 smgrextend(SMgrRelation reln
, ForkNumber forknum
, BlockNumber blocknum
,
504 char *buffer
, bool isTemp
)
506 (*(smgrsw
[reln
->smgr_which
].smgr_extend
)) (reln
, forknum
, blocknum
,
511 * smgrread() -- read a particular block from a relation into the supplied
514 * This routine is called from the buffer manager in order to
515 * instantiate pages in the shared buffer cache. All storage managers
516 * return pages in the format that POSTGRES expects.
519 smgrread(SMgrRelation reln
, ForkNumber forknum
, BlockNumber blocknum
,
522 (*(smgrsw
[reln
->smgr_which
].smgr_read
)) (reln
, forknum
, blocknum
, buffer
);
526 * smgrwrite() -- Write the supplied buffer out.
528 * This is to be used only for updating already-existing blocks of a
529 * relation (ie, those before the current EOF). To extend a relation,
532 * This is not a synchronous write -- the block is not necessarily
533 * on disk at return, only dumped out to the kernel. However,
534 * provisions will be made to fsync the write before the next checkpoint.
536 * isTemp indicates that the relation is a temp table (ie, is managed
537 * by the local-buffer manager). In this case no provisions need be
538 * made to fsync the write before checkpointing.
541 smgrwrite(SMgrRelation reln
, ForkNumber forknum
, BlockNumber blocknum
,
542 char *buffer
, bool isTemp
)
544 (*(smgrsw
[reln
->smgr_which
].smgr_write
)) (reln
, forknum
, blocknum
,
549 * smgrnblocks() -- Calculate the number of blocks in the
553 smgrnblocks(SMgrRelation reln
, ForkNumber forknum
)
555 return (*(smgrsw
[reln
->smgr_which
].smgr_nblocks
)) (reln
, forknum
);
559 * smgrtruncate() -- Truncate supplied relation to the specified number
563 smgrtruncate(SMgrRelation reln
, ForkNumber forknum
, BlockNumber nblocks
,
567 * Get rid of any buffers for the about-to-be-deleted blocks. bufmgr will
568 * just drop them without bothering to write the contents.
570 DropRelFileNodeBuffers(reln
->smgr_rnode
, forknum
, isTemp
, nblocks
);
572 /* Do the truncation */
573 (*(smgrsw
[reln
->smgr_which
].smgr_truncate
)) (reln
, forknum
, nblocks
,
579 * Make an XLOG entry showing the file truncation.
583 xl_smgr_truncate xlrec
;
585 xlrec
.blkno
= nblocks
;
586 xlrec
.rnode
= reln
->smgr_rnode
;
587 xlrec
.forknum
= forknum
;
589 rdata
.data
= (char *) &xlrec
;
590 rdata
.len
= sizeof(xlrec
);
591 rdata
.buffer
= InvalidBuffer
;
594 lsn
= XLogInsert(RM_SMGR_ID
, XLOG_SMGR_TRUNCATE
, &rdata
);
599 * smgrimmedsync() -- Force the specified relation to stable storage.
601 * Synchronously force all previous writes to the specified relation
604 * This is useful for building completely new relations (eg, new
605 * indexes). Instead of incrementally WAL-logging the index build
606 * steps, we can just write completed index pages to disk with smgrwrite
607 * or smgrextend, and then fsync the completed index file before
608 * committing the transaction. (This is sufficient for purposes of
609 * crash recovery, since it effectively duplicates forcing a checkpoint
610 * for the completed index. But it is *not* sufficient if one wishes
611 * to use the WAL log for PITR or replication purposes: in that case
612 * we have to make WAL entries as well.)
614 * The preceding writes should specify isTemp = true to avoid
615 * duplicative fsyncs.
617 * Note that you need to do FlushRelationBuffers() first if there is
618 * any possibility that there are dirty buffers for the relation;
619 * otherwise the sync is not very meaningful.
622 smgrimmedsync(SMgrRelation reln
, ForkNumber forknum
)
624 (*(smgrsw
[reln
->smgr_which
].smgr_immedsync
)) (reln
, forknum
);
629 * PostPrepare_smgr -- Clean up after a successful PREPARE
631 * What we have to do here is throw away the in-memory state about pending
632 * relation deletes. It's all been recorded in the 2PC state file and
633 * it's no longer smgr's job to worry about it.
636 PostPrepare_smgr(void)
638 PendingRelDelete
*pending
;
639 PendingRelDelete
*next
;
641 for (pending
= pendingDeletes
; pending
!= NULL
; pending
= next
)
643 next
= pending
->next
;
644 pendingDeletes
= next
;
645 /* must explicitly free the list entry */
652 * smgrDoPendingDeletes() -- Take care of relation deletes at end of xact.
654 * This also runs when aborting a subxact; we want to clean up a failed
655 * subxact immediately.
658 smgrDoPendingDeletes(bool isCommit
)
660 int nestLevel
= GetCurrentTransactionNestLevel();
661 PendingRelDelete
*pending
;
662 PendingRelDelete
*prev
;
663 PendingRelDelete
*next
;
666 for (pending
= pendingDeletes
; pending
!= NULL
; pending
= next
)
668 next
= pending
->next
;
669 if (pending
->nestLevel
< nestLevel
)
671 /* outer-level entries should not be processed yet */
676 /* unlink list entry first, so we don't retry on failure */
680 pendingDeletes
= next
;
681 /* do deletion if called for */
682 if (pending
->atCommit
== isCommit
)
683 smgr_internal_unlink(pending
->relnode
,
688 /* must explicitly free the list entry */
690 /* prev does not change */
696 * smgrGetPendingDeletes() -- Get a list of relations to be deleted.
698 * The return value is the number of relations scheduled for termination.
699 * *ptr is set to point to a freshly-palloc'd array of RelFileForks.
700 * If there are no relations to be deleted, *ptr is set to NULL.
702 * If haveNonTemp isn't NULL, the bool it points to gets set to true if
703 * there is any non-temp table pending to be deleted; false if not.
705 * Note that the list does not include anything scheduled for termination
706 * by upper-level transactions.
709 smgrGetPendingDeletes(bool forCommit
, RelFileFork
**ptr
, bool *haveNonTemp
)
711 int nestLevel
= GetCurrentTransactionNestLevel();
714 PendingRelDelete
*pending
;
718 *haveNonTemp
= false;
719 for (pending
= pendingDeletes
; pending
!= NULL
; pending
= pending
->next
)
721 if (pending
->nestLevel
>= nestLevel
&& pending
->atCommit
== forCommit
)
729 rptr
= (RelFileFork
*) palloc(nrels
* sizeof(RelFileFork
));
731 for (pending
= pendingDeletes
; pending
!= NULL
; pending
= pending
->next
)
733 if (pending
->nestLevel
>= nestLevel
&& pending
->atCommit
== forCommit
)
735 rptr
->rnode
= pending
->relnode
;
736 rptr
->forknum
= pending
->forknum
;
739 if (haveNonTemp
&& !pending
->isTemp
)
746 * AtSubCommit_smgr() --- Take care of subtransaction commit.
748 * Reassign all items in the pending-deletes list to the parent transaction.
751 AtSubCommit_smgr(void)
753 int nestLevel
= GetCurrentTransactionNestLevel();
754 PendingRelDelete
*pending
;
756 for (pending
= pendingDeletes
; pending
!= NULL
; pending
= pending
->next
)
758 if (pending
->nestLevel
>= nestLevel
)
759 pending
->nestLevel
= nestLevel
- 1;
764 * AtSubAbort_smgr() --- Take care of subtransaction abort.
766 * Delete created relations and forget about deleted relations.
767 * We can execute these operations immediately because we know this
768 * subtransaction will not commit.
771 AtSubAbort_smgr(void)
773 smgrDoPendingDeletes(false);
777 * smgrcommit() -- Prepare to commit changes made during the current
780 * This is called before we actually commit.
787 for (i
= 0; i
< NSmgr
; i
++)
789 if (smgrsw
[i
].smgr_commit
)
790 (*(smgrsw
[i
].smgr_commit
)) ();
795 * smgrabort() -- Clean up after transaction abort.
802 for (i
= 0; i
< NSmgr
; i
++)
804 if (smgrsw
[i
].smgr_abort
)
805 (*(smgrsw
[i
].smgr_abort
)) ();
810 * smgrpreckpt() -- Prepare for checkpoint.
817 for (i
= 0; i
< NSmgr
; i
++)
819 if (smgrsw
[i
].smgr_pre_ckpt
)
820 (*(smgrsw
[i
].smgr_pre_ckpt
)) ();
825 * smgrsync() -- Sync files to disk during checkpoint.
832 for (i
= 0; i
< NSmgr
; i
++)
834 if (smgrsw
[i
].smgr_sync
)
835 (*(smgrsw
[i
].smgr_sync
)) ();
840 * smgrpostckpt() -- Post-checkpoint cleanup.
847 for (i
= 0; i
< NSmgr
; i
++)
849 if (smgrsw
[i
].smgr_post_ckpt
)
850 (*(smgrsw
[i
].smgr_post_ckpt
)) ();
856 smgr_redo(XLogRecPtr lsn
, XLogRecord
*record
)
858 uint8 info
= record
->xl_info
& ~XLR_INFO_MASK
;
860 if (info
== XLOG_SMGR_CREATE
)
862 xl_smgr_create
*xlrec
= (xl_smgr_create
*) XLogRecGetData(record
);
865 reln
= smgropen(xlrec
->rnode
);
866 smgrcreate(reln
, xlrec
->forknum
, false, true);
868 else if (info
== XLOG_SMGR_TRUNCATE
)
870 xl_smgr_truncate
*xlrec
= (xl_smgr_truncate
*) XLogRecGetData(record
);
873 reln
= smgropen(xlrec
->rnode
);
876 * Forcibly create relation if it doesn't exist (which suggests that
877 * it was dropped somewhere later in the WAL sequence). As in
878 * XLogOpenRelation, we prefer to recreate the rel and replay the log
879 * as best we can until the drop is seen.
881 smgrcreate(reln
, xlrec
->forknum
, false, true);
883 /* Can't use smgrtruncate because it would try to xlog */
886 * First, force bufmgr to drop any buffers it has for the to-be-
887 * truncated blocks. We must do this, else subsequent XLogReadBuffer
888 * operations will not re-extend the file properly.
890 DropRelFileNodeBuffers(xlrec
->rnode
, xlrec
->forknum
, false,
893 /* Do the truncation */
894 (*(smgrsw
[reln
->smgr_which
].smgr_truncate
)) (reln
,
899 /* Also tell xlogutils.c about it */
900 XLogTruncateRelation(xlrec
->rnode
, xlrec
->forknum
, xlrec
->blkno
);
903 elog(PANIC
, "smgr_redo: unknown op code %u", info
);
907 smgr_desc(StringInfo buf
, uint8 xl_info
, char *rec
)
909 uint8 info
= xl_info
& ~XLR_INFO_MASK
;
911 if (info
== XLOG_SMGR_CREATE
)
913 xl_smgr_create
*xlrec
= (xl_smgr_create
*) rec
;
915 appendStringInfo(buf
, "file create: %u/%u/%u/%u",
916 xlrec
->rnode
.spcNode
, xlrec
->rnode
.dbNode
,
917 xlrec
->rnode
.relNode
, xlrec
->forknum
);
919 else if (info
== XLOG_SMGR_TRUNCATE
)
921 xl_smgr_truncate
*xlrec
= (xl_smgr_truncate
*) rec
;
923 appendStringInfo(buf
, "file truncate: %u/%u/%u/%u to %u blocks",
924 xlrec
->rnode
.spcNode
, xlrec
->rnode
.dbNode
,
925 xlrec
->rnode
.relNode
, xlrec
->forknum
,
929 appendStringInfo(buf
, "UNKNOWN");