Steve Gibson invents broken SYNcookies
GRC SYN-flood cure is worse than the disease
He dares to call it "GENESIS" (Gibson's ENcryption-Enhanced Spoofing Immunity System). He dares to call it "Beautiful and Perfect." It's the product of "Three Key Innovations" for which he takes credit and which culminate in an "Encrypted Token," which is another way of saying a "SYNcookie", a quite useful thing developed by Dan Bernstein and Eric Schenk back in 1996.
He dares to claim "immunity" from SYN floods. Only Steve Gibson's lame knockoff is dangerously broken.
What it is
A SYN flood is a DoS attack in which server resources, not bandwidth, are stressed. It fakes the initial handshake of a TCP connection with spoofed IPs which the target machine is unable to answer, so the target machine allocates system resources in anticipation of a connection which is never completed. Re-tries and time-outs add up to perhaps three minutes per bogus SYN. A server's capacity to respond to legitimate requests can be devoured in a matter of seconds with very small packets. Only four or five compromised client machines can cripple a server; in this way it's a fiendishly economical attack.
The handshake is simple: a client initiates with a SYN (synchronize) packet; the server replies with a SYN/ACK (synchronize/acknowledge) packet; and the client finalizes with an ACK (acknowledge) packet. If these steps are followed, a TCP connection is established between the two.
GENESIS attempts to negotiate the handshake without allocating system resources until the client's IP can be verified. This is a common-sense approach, essential to SYNcookies as well. But SYNcookies were worked out over time by people who, unlike Gibson, have a solid grasp of TCP/IP and the machines it connects. Even so, it took time and collaboration, and intellectual modesty, to get all the kinks ironed out.
Unfortunately Gibson is so infatuated with the self-created myth of his own genius that he can't be bothered to consult Bernstein and Schenk, or anyone else for that matter, but goes it alone, inspired only by his overweening pride and essential incompetence. Of course his "Beautiful and Perfect" creation is going to be sadly defective. How could it be otherwise?
One Reg reader who wishes to remain anonymous believes that GENESIS is more than a mere failure, but actually worse than no SYN protection at all. It was this person who originally brought the GENESIS project to our attention, and s/he's offered some very insightful observations.
How it's done
Put simply, authenticating a TCP connection request requires the server to encrypt some aspect(s) of the client's and the server's status so as to ensure that the final ACK comes from the same source as the original SYN (pun fully intended).
Data such as the client's ISN (Initial Sequence Number), originating IP and port, MSS (Maximum Segment Size), and the server's IP and port, can be hashed to produce a server ISN which must be available for decoding in the final ACK packet. If the arithmetic fails, the ACK is rejected and no resources are devoted to the bogus connection. If it works out, a connection is made.
Old cookies absolutely need to expire so they can't be reused; and old sequence numbers need to be identifiable so that they don't get mixed up with those belonging to a newer connection. Something unique ('secret') needs to be plugged into the hash so that cookies valid for one server can't be used on another, and so that valid ISNs can't be guessed or bruteforced easily.
Anyone who reads Bernstein and Schenk's correspondence linked above will see that authenticating a SYN request is no trivial matter. There are a number of obstacles, but Gibson manages to overcome only one of them. Yes, he does manage to deal with the problem of disembodied sequence numbers, so that out-of-date numbers aren't carried over to complicate packet reconstruction on a new connection.
But Gibson is silent on the rest of the issues Bernstein and Schenk have labored to solve.
First, he offers no means to cause a cookie (or "Encrypted Token," as he prefers to call it), to expire. A valid cookie can be used to establish a connection. A lot of valid cookies can be used to establish a lot of connections. Perhaps Gibson is unfamiliar with the term 'packet sniffer.' Too bad. We'll just sit back and watch the kiddies gather up zillions of his broken SYNcookies to use against the fools who trust him.
Second, he ignores MSS. It's hard to achieve decent performance without knowing it.
Third, he doesn't use a secret, which means that valid ISNs can be bruteforced and valid ACKs generated -- and abused.
Fourth, he uses RC5, which is slower than MD5 used in SYNcookies -- another performance hit (just in case his gross security sloppiness didn't already frighten you away).
Pants on fire
Gibson dares to pretend that he'd never heard of SYNcookies when he set off in quest of beauty and perfection. "Immediately after I posted the second part of this work to the Web, several participants in the news groups at grc.com reported that similar work had been done before. I was unaware of previous work in this area, and consequently developed my solution independently and without the benefit of any previous work," Steve claims.
I don't believe a word of it. I think he deliberately set out to knock-off SYNcookies and simply failed because the work was too difficult. He's not an übergeek; he just plays one on his Web site.
I did a Google search and turned up more than 7,000 Web pages with the terms 'SYNcookies' or 'SYN cookies'. This guy is hacking TCP, yet he never once encountered a single mention of it?
Impossible. No human being could have his head that far up his own ass -- not even Steve Gibson. ®
Sponsored: Becoming a Pragmatic Security Leader