Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitbfd5c6e

Browse files
committed
WAL-log inplace update before revealing it to other sessions.
A buffer lock won't stop a reader having already checked tuplevisibility. If a vac_update_datfrozenid() and then a crash happenedduring inplace update of a relfrozenxid value, datfrozenxid couldovertake relfrozenxid. That could lead to "could not access status oftransaction" errors. Back-patch to v12 (all supported versions). Inv14 and earlier, this also back-patches the assertion removal fromcommit7fcf2fa.Discussion:https://postgr.es/m/20240620012908.92.nmisch@google.com
1 parent95c5acb commitbfd5c6e

File tree

2 files changed

+46
-16
lines changed

2 files changed

+46
-16
lines changed

‎src/backend/access/heap/README.tuplock

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -203,6 +203,4 @@ Inplace updates create an exception to the rule that tuple data won't change
203203
under a reader holding a pin. A reader of a heap_fetch() result tuple may
204204
witness a torn read. Current inplace-updated fields are aligned and are no
205205
wider than four bytes, and current readers don't need consistency across
206-
fields. Hence, they get by with just fetching each field once. XXX such a
207-
caller may also read a value that has not reached WAL; see
208-
systable_inplace_update_finish().
206+
fields. Hence, they get by with just fetching each field once.

‎src/backend/access/heap/heapam.c

Lines changed: 45 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -6344,13 +6344,18 @@ heap_inplace_update_and_unlock(Relation relation,
63446344
HeapTupleHeaderhtup=oldtup->t_data;
63456345
uint32oldlen;
63466346
uint32newlen;
6347+
char*dst;
6348+
char*src;
63476349

63486350
Assert(ItemPointerEquals(&oldtup->t_self,&tuple->t_self));
63496351
oldlen=oldtup->t_len-htup->t_hoff;
63506352
newlen=tuple->t_len-tuple->t_data->t_hoff;
63516353
if (oldlen!=newlen||htup->t_hoff!=tuple->t_data->t_hoff)
63526354
elog(ERROR,"wrong tuple length");
63536355

6356+
dst= (char*)htup+htup->t_hoff;
6357+
src= (char*)tuple->t_data+tuple->t_data->t_hoff;
6358+
63546359
/*
63556360
* Construct shared cache inval if necessary. Note that because we only
63566361
* pass the new version of the tuple, this mustn't be used for any
@@ -6369,15 +6374,15 @@ heap_inplace_update_and_unlock(Relation relation,
63696374
*/
63706375
PreInplace_Inval();
63716376

6372-
/* NO EREPORT(ERROR) from here till changes are logged */
6373-
START_CRIT_SECTION();
6374-
6375-
memcpy((char*)htup+htup->t_hoff,
6376-
(char*)tuple->t_data+tuple->t_data->t_hoff,
6377-
newlen);
6378-
63796377
/*----------
6380-
* XXX A crash here can allow datfrozenxid() to get ahead of relfrozenxid:
6378+
* NO EREPORT(ERROR) from here till changes are complete
6379+
*
6380+
* Our buffer lock won't stop a reader having already pinned and checked
6381+
* visibility for this tuple. Hence, we write WAL first, then mutate the
6382+
* buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6383+
* checkpoint delay makes that acceptable. With the usual order of
6384+
* changes, a crash after memcpy() and before XLogInsert() could allow
6385+
* datfrozenxid to overtake relfrozenxid:
63816386
*
63826387
* ["D" is a VACUUM (ONLY_DATABASE_STATS)]
63836388
* ["R" is a VACUUM tbl]
@@ -6387,31 +6392,57 @@ heap_inplace_update_and_unlock(Relation relation,
63876392
* D: raise pg_database.datfrozenxid, XLogInsert(), finish
63886393
* [crash]
63896394
* [recovery restores datfrozenxid w/o relfrozenxid]
6395+
*
6396+
* Like in MarkBufferDirtyHint() subroutine XLogSaveBufferForHint(), copy
6397+
* the buffer to the stack before logging. Here, that facilitates a FPI
6398+
* of the post-mutation block before we accept other sessions seeing it.
63906399
*/
6391-
6392-
MarkBufferDirty(buffer);
6400+
Assert((MyProc->delayChkptFlags&DELAY_CHKPT_START)==0);
6401+
START_CRIT_SECTION();
6402+
MyProc->delayChkptFlags |=DELAY_CHKPT_START;
63936403

63946404
/* XLOG stuff */
63956405
if (RelationNeedsWAL(relation))
63966406
{
63976407
xl_heap_inplacexlrec;
6408+
PGAlignedBlockcopied_buffer;
6409+
char*origdata= (char*)BufferGetBlock(buffer);
6410+
Pagepage=BufferGetPage(buffer);
6411+
uint16lower= ((PageHeader)page)->pd_lower;
6412+
uint16upper= ((PageHeader)page)->pd_upper;
6413+
uintptr_tdst_offset_in_block;
6414+
RelFileLocatorrlocator;
6415+
ForkNumberforkno;
6416+
BlockNumberblkno;
63986417
XLogRecPtrrecptr;
63996418

64006419
xlrec.offnum=ItemPointerGetOffsetNumber(&tuple->t_self);
64016420

64026421
XLogBeginInsert();
64036422
XLogRegisterData((char*)&xlrec,SizeOfHeapInplace);
64046423

6405-
XLogRegisterBuffer(0,buffer,REGBUF_STANDARD);
6406-
XLogRegisterBufData(0, (char*)htup+htup->t_hoff,newlen);
6424+
/* register block matching what buffer will look like after changes */
6425+
memcpy(copied_buffer.data,origdata,lower);
6426+
memcpy(copied_buffer.data+upper,origdata+upper,BLCKSZ-upper);
6427+
dst_offset_in_block=dst-origdata;
6428+
memcpy(copied_buffer.data+dst_offset_in_block,src,newlen);
6429+
BufferGetTag(buffer,&rlocator,&forkno,&blkno);
6430+
Assert(forkno==MAIN_FORKNUM);
6431+
XLogRegisterBlock(0,&rlocator,forkno,blkno,copied_buffer.data,
6432+
REGBUF_STANDARD);
6433+
XLogRegisterBufData(0,src,newlen);
64076434

64086435
/* inplace updates aren't decoded atm, don't log the origin */
64096436

64106437
recptr=XLogInsert(RM_HEAP_ID,XLOG_HEAP_INPLACE);
64116438

6412-
PageSetLSN(BufferGetPage(buffer),recptr);
6439+
PageSetLSN(page,recptr);
64136440
}
64146441

6442+
memcpy(dst,src,newlen);
6443+
6444+
MarkBufferDirty(buffer);
6445+
64156446
LockBuffer(buffer,BUFFER_LOCK_UNLOCK);
64166447

64176448
/*
@@ -6424,6 +6455,7 @@ heap_inplace_update_and_unlock(Relation relation,
64246455
*/
64256456
AtInplace_Inval();
64266457

6458+
MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
64276459
END_CRIT_SECTION();
64286460
UnlockTuple(relation,&tuple->t_self,InplaceUpdateTupleLock);
64296461

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp