On Sat, 10 Jun 2000, Alan Cox wrote:
> > Now the question: Are syncookies exactly that without need for an extra
> > firewall? Are they only destination-port and not
...
> Syn cookies will basically cut the effect of a synflood to a load test.
My understanding of the syncookies implementation (in 2.0, at least) is
that it only protects the host which is using it; not the hosts _behind_ it
(ie. acting as a firewall). Is this not the case?
FWIW, I'm researching ways to prevent SYN flooding _though_ a firewall.
One of the approaches I'm interested in hearing feedback on goes something
like this:
1.) The firewall maintains a "Max SYN's per port per sec" value,
configurable in some way.
2.) The firewall tracks SYN's passing through it, watching to see if this
value is ever exceeded (on a per-port basis).
3.) If the value is exceeded, the firewall goes into "SYN flood protection"
mode (for that port), where it begins storing the sender's IP address in a
temp hash table of hosts which require validation. It does _not_ yet try
to validate the sender.*
4.) When a SYN arrives, it is compared to the value in the temp hash table:
if there is not an entry, one is added (as above). If there is an entry,
and it hasn't been validated, a validation request is sent. If there is an
entry and it has already been validated, any more SYN's from this host will
be accepted right away.**
5.) To validate the sender, the firewall sends back a cookie of some
kind*** to validate that the sending system is indeed on the net and has a
listening socket for that connection request.
* Rely on TCP retransmit to cut down on the likelihood of filling the pipe
on outgoing SYN validation attempts.
** Entries in the table expire after just a handful of seconds (20, say) so
that connections like web clients only have a short initial delay for the
first connection, but the rest for that _session_ are unimpeded.
*** Something like an ACK with a bogus seq. number, and the firewall
watches for RST's to update the validation table.
Does this sound reasonable?
-bp
-- # bryan at terran dot org # http://www.terran.org/~bryan- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Jun 15 2000 - 21:00:21 EST