diff --git a/src/core/mount.c b/src/core/mount.c
index dcfbf970d7..b47bd12c96 100644
--- a/src/core/mount.c
+++ b/src/core/mount.c
@@ -2194,9 +2194,6 @@ static int mount_can_start(Unit *u) {
 
         assert(m);
 
-        if (sd_event_source_is_ratelimited(u->manager->mount_event_source))
-                return -EAGAIN;
-
         r = unit_test_start_limit(u);
         if (r < 0) {
                 mount_enter_dead(m, MOUNT_FAILURE_START_LIMIT_HIT);
diff --git a/src/core/unit.c b/src/core/unit.c
index 31c5be4431..180dccc2b2 100644
--- a/src/core/unit.c
+++ b/src/core/unit.c
@@ -1868,6 +1868,10 @@ int unit_start(Unit *u) {
 
         assert(u);
 
+        /* Let's hold off running start jobs for mount units when /proc/self/mountinfo monitor is rate limited. */
+        if (u->type == UNIT_MOUNT && sd_event_source_is_ratelimited(u->manager->mount_event_source))
+                return -EAGAIN;
+
         /* If this is already started, then this will succeed. Note that this will even succeed if this unit
          * is not startable by the user. This is relied on to detect when we need to wait for units and when
          * waiting is finished. */
diff --git a/test/units/testsuite-60.sh b/test/units/testsuite-60.sh
index 239d7b0d4c..8cb3f466b6 100755
--- a/test/units/testsuite-60.sh
+++ b/test/units/testsuite-60.sh
@@ -3,6 +3,56 @@
 set -eux
 set -o pipefail
 
+test_issue_20329() {
+    local tmpdir unit
+    tmpdir="$(mktemp -d)"
+    unit=$(systemd-escape --suffix mount --path "$tmpdir")
+
+    # Set up test mount unit
+    cat > /run/systemd/system/"$unit" <<EOF
+[Mount]
+What=tmpfs
+Where=$tmpdir
+Type=tmpfs
+Options=defaults,nofail
+EOF
+
+    # Start the unit
+    systemctl daemon-reload
+    systemctl start "$unit"
+
+    [[ "$(systemctl show --property SubState --value "$unit")" = "mounted" ]] || {
+        echo >&2 "Test mount \"$unit\" unit isn't mounted"
+        return 1
+    }
+    mountpoint -q "$tmpdir"
+
+    trap 'systemctl stop $unit' RETURN
+
+    # Trigger the mount ratelimiting
+    cd "$(mktemp -d)"
+    mkdir foo
+    for ((i=0;i<50;++i)); do
+        mount --bind foo foo
+        umount foo
+    done
+
+    # Unmount the test mount and start it immediately again via systemd
+    umount "$tmpdir"
+    systemctl start "$unit"
+
+    # Make sure it is seen as mounted by systemd and it actually is mounted
+    [[ "$(systemctl show --property SubState --value "$unit")" = "mounted" ]] || {
+        echo >&2 "Test mount \"$unit\" unit isn't in \"mounted\" state"
+        return 1
+    }
+
+    mountpoint -q "$tmpdir" || {
+        echo >&2 "Test mount \"$unit\" is in \"mounted\" state, actually is not mounted"
+        return 1
+    }
+}
+
 systemd-analyze log-level debug
 systemd-analyze log-target journal
 
@@ -87,6 +137,9 @@ if systemctl list-units -t mount tmp-meow* | grep -q tmp-meow; then
     exit 42
 fi
 
+# test that handling of mount start jobs is delayed when /proc/self/mouninfo monitor is rate limited
+test_issue_20329
+
 systemd-analyze log-level info
 
 echo OK >/testok