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

Commit659903f

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 parentce8c571 commit659903f

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
@@ -6184,13 +6184,18 @@ heap_inplace_update_and_unlock(Relation relation,
61846184
HeapTupleHeaderhtup=oldtup->t_data;
61856185
uint32oldlen;
61866186
uint32newlen;
6187+
char*dst;
6188+
char*src;
61876189

61886190
Assert(ItemPointerEquals(&oldtup->t_self,&tuple->t_self));
61896191
oldlen=oldtup->t_len-htup->t_hoff;
61906192
newlen=tuple->t_len-tuple->t_data->t_hoff;
61916193
if (oldlen!=newlen||htup->t_hoff!=tuple->t_data->t_hoff)
61926194
elog(ERROR,"wrong tuple length");
61936195

6196+
dst= (char*)htup+htup->t_hoff;
6197+
src= (char*)tuple->t_data+tuple->t_data->t_hoff;
6198+
61946199
/*
61956200
* Construct shared cache inval if necessary. Note that because we only
61966201
* pass the new version of the tuple, this mustn't be used for any
@@ -6209,15 +6214,15 @@ heap_inplace_update_and_unlock(Relation relation,
62096214
*/
62106215
PreInplace_Inval();
62116216

6212-
/* NO EREPORT(ERROR) from here till changes are logged */
6213-
START_CRIT_SECTION();
6214-
6215-
memcpy((char*)htup+htup->t_hoff,
6216-
(char*)tuple->t_data+tuple->t_data->t_hoff,
6217-
newlen);
6218-
62196217
/*----------
6220-
* XXX A crash here can allow datfrozenxid() to get ahead of relfrozenxid:
6218+
* NO EREPORT(ERROR) from here till changes are complete
6219+
*
6220+
* Our buffer lock won't stop a reader having already pinned and checked
6221+
* visibility for this tuple. Hence, we write WAL first, then mutate the
6222+
* buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6223+
* checkpoint delay makes that acceptable. With the usual order of
6224+
* changes, a crash after memcpy() and before XLogInsert() could allow
6225+
* datfrozenxid to overtake relfrozenxid:
62216226
*
62226227
* ["D" is a VACUUM (ONLY_DATABASE_STATS)]
62236228
* ["R" is a VACUUM tbl]
@@ -6227,31 +6232,57 @@ heap_inplace_update_and_unlock(Relation relation,
62276232
* D: raise pg_database.datfrozenxid, XLogInsert(), finish
62286233
* [crash]
62296234
* [recovery restores datfrozenxid w/o relfrozenxid]
6235+
*
6236+
* Like in MarkBufferDirtyHint() subroutine XLogSaveBufferForHint(), copy
6237+
* the buffer to the stack before logging. Here, that facilitates a FPI
6238+
* of the post-mutation block before we accept other sessions seeing it.
62306239
*/
6231-
6232-
MarkBufferDirty(buffer);
6240+
Assert((MyProc->delayChkptFlags&DELAY_CHKPT_START)==0);
6241+
START_CRIT_SECTION();
6242+
MyProc->delayChkptFlags |=DELAY_CHKPT_START;
62336243

62346244
/* XLOG stuff */
62356245
if (RelationNeedsWAL(relation))
62366246
{
62376247
xl_heap_inplacexlrec;
6248+
PGAlignedBlockcopied_buffer;
6249+
char*origdata= (char*)BufferGetBlock(buffer);
6250+
Pagepage=BufferGetPage(buffer);
6251+
uint16lower= ((PageHeader)page)->pd_lower;
6252+
uint16upper= ((PageHeader)page)->pd_upper;
6253+
uintptr_tdst_offset_in_block;
6254+
RelFileLocatorrlocator;
6255+
ForkNumberforkno;
6256+
BlockNumberblkno;
62386257
XLogRecPtrrecptr;
62396258

62406259
xlrec.offnum=ItemPointerGetOffsetNumber(&tuple->t_self);
62416260

62426261
XLogBeginInsert();
62436262
XLogRegisterData((char*)&xlrec,SizeOfHeapInplace);
62446263

6245-
XLogRegisterBuffer(0,buffer,REGBUF_STANDARD);
6246-
XLogRegisterBufData(0, (char*)htup+htup->t_hoff,newlen);
6264+
/* register block matching what buffer will look like after changes */
6265+
memcpy(copied_buffer.data,origdata,lower);
6266+
memcpy(copied_buffer.data+upper,origdata+upper,BLCKSZ-upper);
6267+
dst_offset_in_block=dst-origdata;
6268+
memcpy(copied_buffer.data+dst_offset_in_block,src,newlen);
6269+
BufferGetTag(buffer,&rlocator,&forkno,&blkno);
6270+
Assert(forkno==MAIN_FORKNUM);
6271+
XLogRegisterBlock(0,&rlocator,forkno,blkno,copied_buffer.data,
6272+
REGBUF_STANDARD);
6273+
XLogRegisterBufData(0,src,newlen);
62476274

62486275
/* inplace updates aren't decoded atm, don't log the origin */
62496276

62506277
recptr=XLogInsert(RM_HEAP_ID,XLOG_HEAP_INPLACE);
62516278

6252-
PageSetLSN(BufferGetPage(buffer),recptr);
6279+
PageSetLSN(page,recptr);
62536280
}
62546281

6282+
memcpy(dst,src,newlen);
6283+
6284+
MarkBufferDirty(buffer);
6285+
62556286
LockBuffer(buffer,BUFFER_LOCK_UNLOCK);
62566287

62576288
/*
@@ -6264,6 +6295,7 @@ heap_inplace_update_and_unlock(Relation relation,
62646295
*/
62656296
AtInplace_Inval();
62666297

6298+
MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
62676299
END_CRIT_SECTION();
62686300
UnlockTuple(relation,&tuple->t_self,InplaceUpdateTupleLock);
62696301

0 commit comments

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