The code doing addition in that commit is

+                       switch (cmd) {
+                       case TCA_PEDIT_KEY_EX_CMD_SET:
+                               val = tkey->val;
+                               break;
+                       case TCA_PEDIT_KEY_EX_CMD_ADD:
+                               val = (*ptr + tkey->val) & ~tkey->mask;
+                               break;
+                       default:
+                               pr_info("tc filter pedit bad command (%d)\n",
+                                       cmd);
+                               goto bad;
+                       }
+
+                       *ptr = ((*ptr & tkey->mask) ^ val);


Any net-endian field wider than an octet will have the carry between
octets handled wrong on little-endian hosts.  Should we at least
verify that ~mask fits into one octet?

As it is, consider e.g. an attempt to subtract 1 from a 16bit field
at offset 2 in a word.  We want {0,0,0,1} (0x10000000 from host POV)
to turn into 0, so the value to add would be 0xff000000.  Except that
{0, 0, 1, 0} would turn into {0, 0, 1, 0xff} that way, not the
expected {0, 0, 0, 0xff}.

Granted, there's not a lot of wider-than-octet fields where arithmetics
would've made sense, but we probably ought to refuse allowing such
operations.  Especially since on big-endian hosts they will work
just fine until you try to move that over to a little-endian box...

Alternatively, we could do something like
        val = htonl(be32_to_cpup(ptr) + ntohl(tkey->val)) & ~tkey->mask;
but I'm not sure if that's worth doing.  It's not as if there would be
a major overhead, but still...

Comments?

Reply via email to