Make sure that keys are properly removed from hashmap

This is a speculative fix for https://bugzilla.redhat.com/show_bug.cgi?id=1088865.
Even though I cannot find a code path that where this would be
an issue, for consistency, if we assume that cgroup_path might have
been set before we got to unit_deserialize, we should make sure that
the unit is removed from the hashmap before we free the key. This seems
to be the only place where the key could be prematurely freed, leading to
hashmap corruption.

diff --git a/src/core/unit.c b/src/core/unit.c
index 6ac359e31b755daadb280abef5319c5c24002b1c..c4ed9233731a88756e15b855b338bfdaa11a3a12 100644 (file)
--- a/src/core/unit.c
+++ b/src/core/unit.c
@@ -2488,10 +2488,18 @@ int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
                         if (!s)
                                 return -ENOMEM;
 
-                        free(u->cgroup_path);
-                        u->cgroup_path = s;
+                        if (u->cgroup_path) {
+                                void *p;
+
+                                p = hashmap_remove(u->manager->cgroup_unit, u->cgroup_path);
+                                log_info("Removing cgroup_path %s from hashmap (%p)",
+                                         u->cgroup_path, p);
+                                free(u->cgroup_path);
+                        }
 
+                        u->cgroup_path = s;
                         assert(hashmap_put(u->manager->cgroup_unit, s, u) == 1);
+
                         continue;
                 }