As shown in the previous commit, qemu_strtosz_MiB sometimes leaves the
result value untouched (we have to audit further to learn that in that
case, the QAPI generator says that visit_type_NumaOptions() will have
zero-initialized it), and sometimes leaves it with the value of a
partial parse before -EINVAL occurs because of trailing garbage.
Rather than blindly treating any string the user may throw at us as
valid, we should check for parse failures.
Fixes: cc001888 ("numa: fixup parsed NumaNodeOptions earlier", v2.11.0)
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Hanna Czenczek <hreitz@redhat.com>
Message-Id: <
20230522190441.64278-14-eblake@redhat.com>
/* Fix up legacy suffix-less format */
if ((object->type == NUMA_OPTIONS_TYPE_NODE) && object->u.node.has_mem) {
const char *mem_str = qemu_opt_get(opts, "mem");
- qemu_strtosz_MiB(mem_str, NULL, &object->u.node.mem);
+ int ret = qemu_strtosz_MiB(mem_str, NULL, &object->u.node.mem);
+
+ if (ret < 0) {
+ error_setg_errno(&err, -ret, "could not parse memory size '%s'",
+ mem_str);
+ }
}
- set_numa_options(ms, object, &err);
+ if (!err) {
+ set_numa_options(ms, object, &err);
+ }
qapi_free_NumaOptions(object);
if (err) {