[bug report] of: Introduce Device Tree resolve support.

[Dan Carpenter <dan.carpenter@oracle.com>]

Hello Pantelis Antoniou,

The patch 7941b27b16e3: "of: Introduce Device Tree resolve support."
from Jul 4, 2014, leads to the following static checker warning:

	drivers/of/resolver.c:125 update_usages_of_a_phandle_reference()
	error: buffer underflow 'prop->value' 's32min-s32max'

drivers/of/resolver.c
    72  static int update_usages_of_a_phandle_reference(struct device_node *overlay,
    73                  struct property *prop_fixup, phandle phandle)
    74  {
    75          struct device_node *refnode;
    76          struct property *prop;
    77          char *value, *cur, *end, *node_path, *prop_name, *s;
    78          int offset, len;
                    ^^^^^^
    79          int err = 0;
    80  
    81          value = kmemdup(prop_fixup->value, prop_fixup->length, GFP_KERNEL);
    82          if (!value)
    83                  return -ENOMEM;
    84  
    85          /* prop_fixup contains a list of tuples of path:property_name:offset */
    86          end = value + prop_fixup->length;
    87          for (cur = value; cur < end; cur += len + 1) {
    88                  len = strlen(cur);
    89  
    90                  node_path = cur;
    91                  s = strchr(cur, ':');
    92                  if (!s) {
    93                          err = -EINVAL;
    94                          goto err_fail;
    95                  }
    96                  *s++ = '\0';
    97  
    98                  prop_name = s;
    99                  s = strchr(s, ':');
   100                  if (!s) {
   101                          err = -EINVAL;
   102                          goto err_fail;
   103                  }
   104                  *s++ = '\0';
   105  
   106                  err = kstrtoint(s, 10, &offset);
                                               ^^^^^^^
Smatch marks data that we get from the kstrtoint() as untrusted.

   107                  if (err)
   108                          goto err_fail;
   109  
   110                  refnode = __of_find_node_by_full_path(of_node_get(overlay), node_path);
   111                  if (!refnode)
   112                          continue;
   113  
   114                  for_each_property_of_node(refnode, prop) {
   115                          if (!of_prop_cmp(prop->name, prop_name))
   116                                  break;
   117                  }
   118                  of_node_put(refnode);
   119  
   120                  if (!prop) {
   121                          err = -ENOENT;
   122                          goto err_fail;
   123                  }
   124  
   125                  *(__be32 *)(prop->value + offset) = cpu_to_be32(phandle);
                                    ^^^^^^^^^^^^^^^^^^^^
So then it complains about this...

We probably trust this data and are fine with it writing to where ever
but it would be nice for reviewers to prevent the array
underflow/overflow warning.  I'm not even sure how big prop->value is
though so I don't know what the bounds check should be.

   126          }
   127  
   128  err_fail:
   129          kfree(value);
   130          return err;

regards,
dan carpenter