Re: Preventing Denial of Service Attack In IPC Serialization
On Jun 12, 11:29 am, "Joe" <j...@junk.com> wrote:
You seem to want to use the same syntax to serialize to everything.
Although
I do not understand all the issues that have been brought up about the
coupling of the serialization and transmission concepts, you could do
something similar to the following to "have your cake and eat it too";
class Socket:public Archive{};
template<typename T> Socket& operator<<(Socket& socket, const T& t){
ostringstream oss();
oss << T;
socket.send(oss.str().size());
socket.send(oss.str());
return s;
};
template<typename T> Socket& operator>>(Socket& socket, T& t){
string sBuf;
size_t n;
socket.get(&n);
sBuf.resize(n);
socket.get(&sBuf);
istringstream iss(sBuf);
iss >> T;
return s;
}
The problem is "n". If an attacker at the other end of the connection
defines n to be, for example, 2^32, then you have a problem.
You could use the same syntax. There is ample opportunity to some error
checking in both operator<< and >> functions. you could also get your
download string in chucks as well (not shown).
Chunks are nice, but breaking-into-chunks was have been taken for
granted from the beginning. The maximum size of Ethernet payload is
1500 bytes, so naturally, no one is sending 1MB packets. The problem
is the reassembly phase of the object at the receiver end. Without
security, the receiver is left vulnerable, known that, at any moment,
it will be induced to consume massive amounts of memory from the free-
store.
Note that any attempt to mitigate the problem put putting "checks" in
the serialization code to control somewhat the amount of memory
allocated will not work. The issue becomes the word "somewhat". What
is "somewhat?" 1KB? 16KB? 1MB? It's like a doctor being told to prep
drug for medically-induced coma for incoming patient without telling
him anything about the patient. Any preconceived dosage (limit on
memory allocation) would be "unreasonable", not to mention that the
serialization code would become very ugly, very fast.
This seem to separate serialization and transmission while allowing you to
use the same syntax. Yes/No ?
Yes but fundamental problem still persists.
-Le Chaud Lapin-
--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]