Fix another longstanding problem in copy_relation_data: it was blithely

tglsfdc · tglsfdc · commit 8e21bf525046 · 2010-07-29T19:24:05.000Z
assuming that a local char[] array would be aligned on at least a word
boundary.  There are architectures on which that is pretty much guaranteed to
NOT be the case ... and those arches also don't like non-aligned memory
accesses, meaning that log_newpage() would crash if it ever got invoked.
Even on Intel-ish machines there's a potential for a large performance penalty
from doing I/O to an inadequately aligned buffer.  So palloc it instead.

Backpatch to 8.0 --- 7.4 doesn't have this code.
diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/commands/tablecmds.c,v 1.142.4.13 2010/07/29 16:15:47 rhaas Exp $
+ *	  $PostgreSQL: pgsql/src/backend/commands/tablecmds.c,v 1.142.4.14 2010/07/29 19:24:05 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -5729,11 +5729,11 @@ static void
 copy_relation_data(Relation rel, SMgrRelation dst)
 {
 	SMgrRelation src;
+	char	   *buf;
+	Page		page;
 	bool		use_wal;
 	BlockNumber nblocks;
 	BlockNumber blkno;
-	char		buf[BLCKSZ];
-	Page		page = (Page) buf;
 
 	/*
 	 * Since we copy the data directly without looking at the shared
@@ -5744,6 +5744,15 @@ copy_relation_data(Relation rel, SMgrRelation dst)
 	 */
 	FlushRelationBuffers(rel, 0);
 
+	/*
+	 * palloc the buffer so that it's MAXALIGN'd.  If it were just a local
+	 * char[] array, the compiler might align it on any byte boundary, which
+	 * can seriously hurt transfer speed to and from the kernel; not to
+	 * mention possibly making PageSetLSN fail.
+	 */
+	buf = (char *) palloc(BLCKSZ);
+	page = (Page) buf;
+
 	/*
 	 * We need to log the copied data in WAL iff WAL archiving is enabled
 	 * AND it's not a temp rel.
@@ -5807,6 +5816,8 @@ copy_relation_data(Relation rel, SMgrRelation dst)
 		smgrwrite(dst, blkno, buf, true);
 	}
 
+	pfree(buf);
+
 	/*
 	 * If the rel isn't temp, we must fsync it down to disk before it's
 	 * safe to commit the transaction.	(For a temp rel we don't care
