Branch b1_6

b=15594
i=adilger, shadow

The dd has a change of behaviour in 2.6.22 kernel. It might return
a partial read from read() system-call in some case.

lustre/tests/sanityN.sh

@@ -201,17 +201,19 @@ test_10a() {
 run_test 10a "write of file with sub-page size on multiple mounts "
 
 test_10b() {
-	yes "R" | dd of=$DIR1/f10b bs=3k count=1 || error "dd $DIR1"
+	# create a seed file
+	yes "R" | head -c 4000 >$TMP/f10b-seed
+	dd if=$TMP/f10b-seed of=$DIR1/f10b bs=3k count=1 || error "dd $DIR1"
 
 	truncate $DIR1/f10b 4096 || error "truncate 4096"
 
 	dd if=$DIR2/f10b of=$TMP/f10b-lustre bs=4k count=1 || error "dd $DIR2"
 
 	# create a test file locally to compare
-	yes "R" | dd of=$TMP/f10b bs=3k count=1 || error "dd random"
+	dd if=$TMP/f10b-seed of=$TMP/f10b bs=3k count=1 || error "dd random"
 	truncate $TMP/f10b 4096 || error "truncate 4096"
 	cmp $TMP/f10b $TMP/f10b-lustre || error "file miscompare"
-	rm $TMP/f10b $TMP/f10b-lustre
+	rm $TMP/f10b $TMP/f10b-lustre $TMP/f10b-seed
 }
 run_test 10b "write of file with sub-page size on multiple mounts "