database/postgres
database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-04-22 23:02:54 +03:00
Code

Fix double-XLogBeginInsert call in GIN page splits.

Browse Source 
If data checksums or wal_log_hints is on, and a GIN page is split, the code
to find a new, empty, block was called after having already called
XLogBeginInsert(). That causes an assertion failure or PANIC, if finding the
new block involves updating a FSM page that had not been modified since last
checkpoint, because that update is WAL-logged, which calls XLogBeginInsert
again. Nested XLogBeginInsert calls are not supported.

To fix, rearrange GIN code so that XLogBeginInsert is called later, after
finding the victim buffers.

Reported by Jeff Janes.
This commit is contained in:
Heikki Linnakangas 2015-06-28 21:59:29 +03:00
parent b36805f3c5
commit a45c70acf3
3 changed files with 11 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
src/backend/access/gin/ginbtree.c
View File

@ -358,20 +358,15 @@ ginPlaceToPage(GinBtree btree, GinBtreeStack *stack,
* placeToPage can register some data to the WAL record.
*
* If placeToPage returns INSERTED, placeToPage has already called
* START_CRIT_SECTION(), and we're responsible for calling
* END_CRIT_SECTION. When it returns INSERTED, it is also responsible for
* registering any data required to replay the operation with
* XLogRegisterData(0, ...). It may only add data to block index 0; the
* main data of the WAL record is reserved for this function.
* START_CRIT_SECTION() and XLogBeginInsert(), and registered any data
* required to replay the operation, in block index 0. We're responsible
* for filling in the main data portion of the WAL record, calling
* XLogInsert(), and END_CRIT_SECTION.
*
* If placeToPage returns SPLIT, we're wholly responsible for WAL logging.
* Splits happen infrequently, so we just make a full-page image of all
* the pages involved.
*/
if (RelationNeedsWAL(btree->index))
XLogBeginInsert();
rc = btree->placeToPage(btree, stack->buffer, stack,
insertdata, updateblkno,
&newlpage, &newrpage);
@ -558,6 +553,8 @@ ginPlaceToPage(GinBtree btree, GinBtreeStack *stack,
{
XLogRecPtr recptr;
XLogBeginInsert();
/*
* We just take full page images of all the split pages. Splits
* are uncommon enough that it's not worth complicating the code

4
src/backend/access/gin/gindatapage.c
View File

@ -600,7 +600,10 @@ dataPlaceToPageLeaf(GinBtree btree, Buffer buf, GinBtreeStack *stack,
*/
MemoryContextSwitchTo(oldCxt);
if (RelationNeedsWAL(btree->index))
{
XLogBeginInsert();
registerLeafRecompressWALData(buf, leaf);
}
START_CRIT_SECTION();
dataPlaceToPageLeafRecompress(buf, leaf);
@ -1120,6 +1123,7 @@ dataPlaceToPageInternal(GinBtree btree, Buffer buf, GinBtreeStack *stack,
data.offset = off;
data.newitem = *pitem;
XLogBeginInsert();
XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) &data,
sizeof(ginxlogInsertDataInternal));

1
src/backend/access/gin/ginentrypage.c
View File

@ -557,6 +557,7 @@ entryPlaceToPage(GinBtree btree, Buffer buf, GinBtreeStack *stack,
data.isDelete = insertData->isDelete;
data.offset = off;
XLogBeginInsert();
XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) &data,
offsetof(ginxlogInsertEntry, tuple));