Avoid allocations in critical sections.

hlinnaka · hlinnaka · commit 7d1e0e8d7a96 · 2014-04-04T13:22:45.000+03:00
If a palloc in a critical section fails, it becomes a PANIC.
diff --git a/src/backend/access/nbtree/nbtinsert.c b/src/backend/access/nbtree/nbtinsert.c
@@ -1839,8 +1839,10 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	BTPageOpaque rootopaque;
 	ItemId		itemid;
 	IndexTuple	item;
-	Size		itemsz;
-	IndexTuple	new_item;
+	IndexTuple	left_item;
+	Size		left_item_sz;
+	IndexTuple	right_item;
+	Size		right_item_sz;
 	Buffer		metabuf;
 	Page		metapg;
 	BTMetaPageData *metad;
@@ -1859,6 +1861,26 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	metapg = BufferGetPage(metabuf);
 	metad = BTPageGetMeta(metapg);
 
+	/*
+	 * Create downlink item for left page (old root).  Since this will be the
+	 * first item in a non-leaf page, it implicitly has minus-infinity key
+	 * value, so we need not store any actual key in it.
+	 */
+	left_item_sz = sizeof(IndexTupleData);
+	left_item = (IndexTuple) palloc(left_item_sz);
+	left_item->t_info = left_item_sz;
+	ItemPointerSet(&(left_item->t_tid), lbkno, P_HIKEY);
+
+	/*
+	 * Create downlink item for right page.  The key for it is obtained from
+	 * the "high key" position in the left page.
+	 */
+	itemid = PageGetItemId(lpage, P_HIKEY);
+	right_item_sz = ItemIdGetLength(itemid);
+	item = (IndexTuple) PageGetItem(lpage, itemid);
+	right_item = CopyIndexTuple(item);
+	ItemPointerSet(&(right_item->t_tid), rbkno, P_HIKEY);
+
 	/* NO EREPORT(ERROR) from here till newroot op is logged */
 	START_CRIT_SECTION();
 
@@ -1876,16 +1898,6 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	metad->btm_fastroot = rootblknum;
 	metad->btm_fastlevel = rootopaque->btpo.level;
 
-	/*
-	 * Create downlink item for left page (old root).  Since this will be the
-	 * first item in a non-leaf page, it implicitly has minus-infinity key
-	 * value, so we need not store any actual key in it.
-	 */
-	itemsz = sizeof(IndexTupleData);
-	new_item = (IndexTuple) palloc(itemsz);
-	new_item->t_info = itemsz;
-	ItemPointerSet(&(new_item->t_tid), lbkno, P_HIKEY);
-
 	/*
 	 * Insert the left page pointer into the new root page.  The root page is
 	 * the rightmost page on its level so there is no "high key" in it; the
@@ -1894,32 +1906,20 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	 * Note: we *must* insert the two items in item-number order, for the
 	 * benefit of _bt_restore_page().
 	 */
-	if (PageAddItem(rootpage, (Item) new_item, itemsz, P_HIKEY,
+	if (PageAddItem(rootpage, (Item) left_item, left_item_sz, P_HIKEY,
 					false, false) == InvalidOffsetNumber)
 		elog(PANIC, "failed to add leftkey to new root page"
 			 " while splitting block %u of index \"%s\"",
 			 BufferGetBlockNumber(lbuf), RelationGetRelationName(rel));
-	pfree(new_item);
-
-	/*
-	 * Create downlink item for right page.  The key for it is obtained from
-	 * the "high key" position in the left page.
-	 */
-	itemid = PageGetItemId(lpage, P_HIKEY);
-	itemsz = ItemIdGetLength(itemid);
-	item = (IndexTuple) PageGetItem(lpage, itemid);
-	new_item = CopyIndexTuple(item);
-	ItemPointerSet(&(new_item->t_tid), rbkno, P_HIKEY);
 
 	/*
 	 * insert the right page pointer into the new root page.
 	 */
-	if (PageAddItem(rootpage, (Item) new_item, itemsz, P_FIRSTKEY,
+	if (PageAddItem(rootpage, (Item) right_item, right_item_sz, P_FIRSTKEY,
 					false, false) == InvalidOffsetNumber)
 		elog(PANIC, "failed to add rightkey to new root page"
 			 " while splitting block %u of index \"%s\"",
 			 BufferGetBlockNumber(lbuf), RelationGetRelationName(rel));
-	pfree(new_item);
 
 	MarkBufferDirty(rootbuf);
 	MarkBufferDirty(metabuf);
@@ -1967,6 +1967,9 @@ _bt_newroot(Relation rel, Buffer lbuf, Buffer rbuf)
 	/* done with metapage */
 	_bt_relbuf(rel, metabuf);
 
+	pfree(left_item);
+	pfree(right_item);
+
 	return rootbuf;
 }
 
diff --git a/src/backend/access/spgist/spgdoinsert.c b/src/backend/access/spgist/spgdoinsert.c
@@ -121,7 +121,8 @@ cmpOffsetNumbers(const void *a, const void *b)
  *
  * NB: this is used during WAL replay, so beware of trying to make it too
  * smart.  In particular, it shouldn't use "state" except for calling
- * spgFormDeadTuple().
+ * spgFormDeadTuple().  This is also used in a critical section, so no
+ * pallocs either!
  */
 void
 spgPageIndexMultiDelete(SpGistState *state, Page page,
@@ -130,7 +131,7 @@ spgPageIndexMultiDelete(SpGistState *state, Page page,
 						BlockNumber blkno, OffsetNumber offnum)
 {
 	OffsetNumber firstItem;
-	OffsetNumber *sortednos;
+	OffsetNumber sortednos[MaxIndexTuplesPerPage];
 	SpGistDeadTuple tuple = NULL;
 	int			i;
 
@@ -144,7 +145,6 @@ spgPageIndexMultiDelete(SpGistState *state, Page page,
 	 * replacement tuples.)  However, we must not scribble on the caller's
 	 * array, so we have to make a copy.
 	 */
-	sortednos = (OffsetNumber *) palloc(sizeof(OffsetNumber) * nitems);
 	memcpy(sortednos, itemnos, sizeof(OffsetNumber) * nitems);
 	if (nitems > 1)
 		qsort(sortednos, nitems, sizeof(OffsetNumber), cmpOffsetNumbers);
@@ -172,8 +172,6 @@ spgPageIndexMultiDelete(SpGistState *state, Page page,
 		else if (tupstate == SPGIST_PLACEHOLDER)
 			SpGistPageGetOpaque(page)->nPlaceholder++;
 	}
-
-	pfree(sortednos);
 }
 
 /*
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
@@ -2449,6 +2449,7 @@ XLogFileInit(uint32 log, uint32 seg,
 {
 	char		path[MAXPGPATH];
 	char		tmppath[MAXPGPATH];
+	char		zbuffer_raw[BLCKSZ + MAXIMUM_ALIGNOF];
 	char	   *zbuffer;
 	uint32		installed_log;
 	uint32		installed_seg;
@@ -2506,11 +2507,11 @@ XLogFileInit(uint32 log, uint32 seg,
 	 * fdatasync(2) or O_DSYNC will be sufficient to sync future writes to the
 	 * log file.
 	 *
-	 * Note: palloc zbuffer, instead of just using a local char array, to
-	 * ensure it is reasonably well-aligned; this may save a few cycles
-	 * transferring data to the kernel.
+	 * Note: ensure the buffer is reasonably well-aligned; this may save a few
+	 * cycles transferring data to the kernel.
 	 */
-	zbuffer = (char *) palloc0(XLOG_BLCKSZ);
+	zbuffer = (char *) MAXALIGN(zbuffer_raw);
+	memset(zbuffer, 0, BLCKSZ);
 	for (nbytes = 0; nbytes < XLogSegSize; nbytes += XLOG_BLCKSZ)
 	{
 		errno = 0;
@@ -2530,7 +2531,6 @@ XLogFileInit(uint32 log, uint32 seg,
 					 errmsg("could not write to file \"%s\": %m", tmppath)));
 		}
 	}
-	pfree(zbuffer);
 
 	if (pg_fsync(fd) != 0)
 		ereport(ERROR,
diff --git a/src/backend/storage/page/bufpage.c b/src/backend/storage/page/bufpage.c
@@ -15,6 +15,7 @@
 #include "postgres.h"
 
 #include "access/htup.h"
+#include "access/itup.h"
 
 
 /* ----------------------------------------------------------------
@@ -362,8 +363,6 @@ PageRepairFragmentation(Page page)
 	Offset		pd_lower = ((PageHeader) page)->pd_lower;
 	Offset		pd_upper = ((PageHeader) page)->pd_upper;
 	Offset		pd_special = ((PageHeader) page)->pd_special;
-	itemIdSort	itemidbase,
-				itemidptr;
 	ItemId		lp;
 	int			nline,
 				nstorage,
@@ -413,10 +412,11 @@ PageRepairFragmentation(Page page)
 		((PageHeader) page)->pd_upper = pd_special;
 	}
 	else
-	{							/* nstorage != 0 */
+	{
 		/* Need to compact the page the hard way */
-		itemidbase = (itemIdSort) palloc(sizeof(itemIdSortData) * nstorage);
-		itemidptr = itemidbase;
+		itemIdSortData itemidbase[MaxHeapTuplesPerPage];
+		itemIdSort	itemidptr = itemidbase;
+
 		totallen = 0;
 		for (i = 0; i < nline; i++)
 		{
@@ -461,8 +461,6 @@ PageRepairFragmentation(Page page)
 		}
 
 		((PageHeader) page)->pd_upper = upper;
-
-		pfree(itemidbase);
 	}
 
 	/* Set hint bit for PageAddItem */
@@ -711,8 +709,8 @@ PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems)
 	Offset		pd_lower = phdr->pd_lower;
 	Offset		pd_upper = phdr->pd_upper;
 	Offset		pd_special = phdr->pd_special;
-	itemIdSort	itemidbase,
-				itemidptr;
+	itemIdSortData	itemidbase[MaxIndexTuplesPerPage];
+	itemIdSort	itemidptr;
 	ItemId		lp;
 	int			nline,
 				nused;
@@ -724,6 +722,8 @@ PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems)
 	int			nextitm;
 	OffsetNumber offnum;
 
+	Assert(nitems < MaxIndexTuplesPerPage);
+
 	/*
 	 * If there aren't very many items to delete, then retail
 	 * PageIndexTupleDelete is the best way.  Delete the items in reverse
@@ -758,7 +758,6 @@ PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems)
 	 * still validity-checking.
 	 */
 	nline = PageGetMaxOffsetNumber(page);
-	itemidbase = (itemIdSort) palloc(sizeof(itemIdSortData) * nline);
 	itemidptr = itemidbase;
 	totallen = 0;
 	nused = 0;
@@ -824,6 +823,4 @@ PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems)
 
 	phdr->pd_lower = SizeOfPageHeaderData + nused * sizeof(ItemIdData);
 	phdr->pd_upper = upper;
-
-	pfree(itemidbase);
 }

Original file line number	Diff line number	Diff line change
`@@ -2449,6 +2449,7 @@ XLogFileInit(uint32 log, uint32 seg,`
`2449`	`2449`	`{`
`2450`	`2450`	`char path[MAXPGPATH];`
`2451`	`2451`	`char tmppath[MAXPGPATH];`
	`2452`	`+ char zbuffer_raw[BLCKSZ + MAXIMUM_ALIGNOF];`
`2452`	`2453`	`char *zbuffer;`
`2453`	`2454`	`uint32 installed_log;`
`2454`	`2455`	`uint32 installed_seg;`
`@@ -2506,11 +2507,11 @@ XLogFileInit(uint32 log, uint32 seg,`
`2506`	`2507`	`* fdatasync(2) or O_DSYNC will be sufficient to sync future writes to the`
`2507`	`2508`	`* log file.`
`2508`	`2509`	`*`
`2509`		`- * Note: palloc zbuffer, instead of just using a local char array, to`
`2510`		`- * ensure it is reasonably well-aligned; this may save a few cycles`
`2511`		`- * transferring data to the kernel.`
	`2510`	`+ * Note: ensure the buffer is reasonably well-aligned; this may save a few`
	`2511`	`+ * cycles transferring data to the kernel.`
`2512`	`2512`	`*/`
`2513`		`- zbuffer = (char *) palloc0(XLOG_BLCKSZ);`
	`2513`	`+ zbuffer = (char *) MAXALIGN(zbuffer_raw);`
	`2514`	`+ memset(zbuffer, 0, BLCKSZ);`
`2514`	`2515`	`for (nbytes = 0; nbytes < XLogSegSize; nbytes += XLOG_BLCKSZ)`
`2515`	`2516`	`{`
`2516`	`2517`	`errno = 0;`
`@@ -2530,7 +2531,6 @@ XLogFileInit(uint32 log, uint32 seg,`
`2530`	`2531`	`errmsg("could not write to file \"%s\": %m", tmppath)));`
`2531`	`2532`	`}`
`2532`	`2533`	`}`
`2533`		`- pfree(zbuffer);`
`2534`	`2534`
`2535`	`2535`	`if (pg_fsync(fd) != 0)`
`2536`	`2536`	`ereport(ERROR,`