tests/valgrind: improve handling for fork() and exec()

- use trace-children=yes.  (This actually means "trace exec"; the
  valgrind man page even admits that the option is badly named!)

- allow our tests to only check the parent pid

In more detail, in our tests/??-*.sh, we would set up:

    scenario_cmd() {
        ...
        c_valgrind_cmd=$(valgrind_setup_cmd "valgrind-parent")
        ${c_valgrind_cmd} ./test_something_with_forks

If we have two pids from a single binary, then our valgrind test will
assume that we're using daemonize() and ignore any leaks from the parent
pid.  (This is exactly what we want for daemonize().)

However, if we use fork() but keep the parent, then we want the
opposite: we should check for leaks in the parent, and ignore leaks from
the children.

We need to explicitly add something to the log filename beacuse if we
didn't, then a valgrind check with two pids wouldn't know whether to
complain about leaks in the parent or child.

tests/shared_valgrind_functions.sh

@@ -10,9 +10,9 @@ set -o noclobber -o nounset
 # - valgrind_init():
 #   Clear previous valgrind output, and prepare for running valgrind tests
 #   (if applicable).
-# - valgrind_setup_cmd():
+# - valgrind_setup_cmd(str):
 #   Set up the valgrind command if ${USE_VALGRIND} is greater than or equal to
-#   ${valgrind_min}.
+#   ${valgrind_min}.  If ${str} is not blank, include it in the log filename.
 # - valgrind_check_basenames(exitfile):
 #   Check for any memory leaks recorded in valgrind logfiles associated with a
 #   test exitfile.  Return the filename if there's a leak; otherwise return an
@@ -203,6 +203,7 @@ _valgrind_ensure_suppression() {
 		printf "\n" | (valgrind					\
 		    --leak-check=full --show-leak-kinds=all		\
 		    --gen-suppressions=all				\
+		    --trace-children=yes				\
 		    --suppressions="${valgrind_suppressions}"		\
 		    --log-file="${this_valgrind_supp}"			\
 		    "${potential_memleaks_binary}"			\
@@ -224,17 +225,27 @@ _valgrind_ensure_suppression() {
 
 ## valgrind_setup_cmd ():
 # Set up the valgrind command if ${USE_VALGRIND} is greater than or equal to
-# ${valgrind_min}.
+# ${valgrind_min}.  If ${str} is not blank, include it in the log filename.
 valgrind_setup_cmd() {
+	str=${1:-}
+
 	# Bail if we don't want to use valgrind for this check.
 	if [ "${USE_VALGRIND}" -lt "${c_valgrind_min}" ]; then
 		return
 	fi
 
-	val_logfilename="${s_val_basename}-${c_count_str}-%p.log"
+	# Set up the log filename.
+	if [ -n "${str}" ]; then
+		val_logfilename="${s_val_basename}-${c_count_str}-${str}-%p.log"
+	else
+		val_logfilename="${s_val_basename}-${c_count_str}-%p.log"
+	fi
+
+	# Set up valgrind command.
 	c_valgrind_cmd="valgrind \
 		--log-file=${val_logfilename} \
 		--track-fds=yes \
+		--trace-children=yes \
 		--leak-check=full --show-leak-kinds=all \
 		--errors-for-leak-kinds=all \
 		--suppressions=${valgrind_suppressions}"
@@ -345,6 +356,16 @@ valgrind_check_basenames() {
 		return
 	fi
 
+	# If the valgrind logfiles contain "-valgrind-parent-", then we only
+	# want to check the parent (the lowest pid).
+	for logfile in ${logfiles} ; do
+		if [ "${logfile#*-valgrind-parent-}" != "${logfile}" ]; then
+			# Only check the parent
+			valgrind_check_logfile "${logfile}"
+			return "$?"
+		fi
+	done
+
 	# If there's two files, there's a fork() -- likely within
 	# daemonize() -- so only pay attention to the child.
 	if [ "${num_logfiles}" -eq "2" ]; then