Download - Secure Protocols and VPNs
What we’ll cover
• This lecture:– network layering revision– “secure channel” concept– substantial example: IPSec
• Next lecture:– substantial example 1: TLS/SSL– substantial example 2: SSH– summaries and random rants
Network layers
• You heard this in Lecture 1!• Simplified Internet model:
applicationtransportnetworklink
Network layersapplication datale
nTCP headerport, seqnum,
SYN, ACK, FIN, …
application datalenTCP header
port, seqnum,SYN, ACK, FIN, …
lenIP header
srcIP, destIP,TTL, pktID, …
application datalenTCP header
port, seqnum,SYN, ACK, FIN, …
lenIP header
srcIP, destIP,TTL, pktID, …
Ethernet headerether-addr, checksum, … le
n
application datalenTCP header
port, seqnum,SYN, ACK, FIN, …
lenIP header’
srcIP, destIP,TTL-n, pktID, …
X.25 headerVirtual-circuit-ID, … le
n
Application layer: lots of protocols, e.g. HTTP carrying Web traffic, SMTP carrying email, NNTP carrying News, …
Transport layer: TCP - reliable connections (sequence nums, retransmission), carries HTTP, FTP, Telnet, …; UDP – unreliable datagrams, e.g. streaming audio/video
Network (IP) layer: “is” Internet. Carries: TCP, UDP, ICMP (“ping”, router control), …
Link layer: carries IP (and others, e.g. Appletalk, NetBUI, IPX, ARP, ...): hop-by-hop LAN.Examples: IEEE802.3, PPP, RF LAN
Where shall we “put security”?
• Link level:covers all traffic on that link, e.g. RF only one hop
• Network (IP) level:covers “all” traffic, end-to-endtransparent to applications little application control “unnatural”, since IP is stateless packets
but channel is stateful
Where shall we “put security”?
• Transport (TCP) level:end-to-endapps can control when it’s used apps must be modified (unless proxied)
• Application level:can be tuned to payload requirements no “leverage” - must rework for every
app
What “security” are we providing?
• A “secure channel”, typically offering– Origin authentication (but of what: OS? App?
User?)
– Integrity– Confidentiality– …
• Not:– Non-repudiation– Any services once data received
Crypto primitives used
• Symmetric ciphers• Asymmetric ciphers & signatures• (Keyed) hash functions• (Keyed) pseudo-random functions• Key agreement protocols: mainly
DH
Typical goals
• At least one party authenticated• Shared secret established• MAC & bulk cipher keys derived
from shared secret• Further traffic “protected”• Optional: session re-use, rekeying,
…
IPSec: overview
• Network-level: all IP datagrams covered• Mandatory for next-generation IP (v6),
optional for current-generation (v4)• Authentication-only or confidentiality too• Two “modes”
– “transport” mode: for IPSec-aware hosts as endpoints
– “tunnel” mode: for IPSec-unaware hosts, established by intermediate gateways or host OS
References: RFC2401..2412; FreeSWAN
Authentication: AH
• Authenticates whole payload and most of header– vitally, covers source IP address– omits only fields modified in transit
• e.g. TTL/hop-limit, fragmentation fields, some options
Encryption + auth: ESP
• Encrypts and optionally authenticates payload, but not IP header
• Combine with AH for “full” conf+auth• Use alone for payload conf+auth• There are both engineering and
political reasons for their separate existence!
AH & ESP, Tunnel & Transport
application datalenTCP header
port, seqnum,SYN, ACK, FIN, …
lenIP header
srcIP, destIP,TTL, pktID, …
application datalenTCP headerlenIP header
srcIP, destIP,TTL, pktID, …
AHSPI, seqnum,
MAC, …
MAC scope
application datalenTCP headerlenIP header
srcIP, destIP,TTL, pktID, …
ESPheader:
SPI,seqnum
MAC scope
ESPtrailer: pad,padlen, …
encryption scope
application datalenTCP headerlenIP header
srcIP, destIP,TTL, pktID, …
ESPheader:
SPI, seqnum
MAC scope
ESPauth: MAC
application datalenTCP headerlenInner IP header
host-srcIP, host-destIP,TTL, pktID, …
AHSPI, seqnum,
MAC. …
MAC scope
Outer IP headergw-srcIP, gw-destIP,
TTL, pktID, …
len
application datalenTCP headerlenInner IP header
host-srcIP, host-destIP,TTL, pktID, …
AHSPI,seqnum,
MAC, …
Inner AH MAC scope
Outer IP headergw-srcIP, gw-destIP,
TTL, pktID, …
len ESP
trailer: pad,padlen…
Outer ESP MAC scope
ESPheader:
SPI,seqnum
encryption scope
Original unprotected datagram
AH in transport mode
ESP (conf and auth)in transport mode
ESP auth-onlyin transport mode
AH in tunnel mode
ESP I(conf and auth) in tunnel mode
carrying AH in transport mode
ESPauth: MAC
ESPauth: MAC
Why so many combinations!?
• To support different VPN arrangements, to meet different security and deployment-practicality requirements
Simple transport-mode usage:
host-to-host(no singing, please)
app datalenTCPlenIP
app datalenTCPlenIP
app datalenTCPlenIP lenAH lenESP ESP
app datalenTCPlenIP lenAH lenESP ESP
Without transport-mode IPSec
With transport-mode IPSec
Simple tunnel-mode usage:
gateway-to-gateway
app datalenTCPlenIP
host-to-host app datale
nTCPlenIP
host-to-host
app datalenTCPlenIP
gw-to-gw
lenESP lenIP
host-to-host
ESP app datalenTCPlenIP
gw-to-gw
lenESP lenIP
host-to-host
ESP
Other combinations for other requirements
application datalenTCP headerlenIP header
srcIP, destIP,TTL, pktID, …
AHSPI, seqnum,
MAC, …
MAC scope
application datalenTCP headerlenIP header
srcIP, destIP,TTL, pktID, …
ESPheader:
SPI,seqnum
MAC scope
ESPtrailer: pad,padlen, …
encryption scope
application datalenTCP headerlenIP header
srcIP, destIP,TTL, pktID, …
ESPheader:
SPI, seqnum
MAC scope
ESPauth: MAC
application datalenTCP headerlenInner IP header
host-srcIP, host-destIP,TTL, pktID, …
AHSPI, seqnum,
MAC. …
MAC scope
Outer IP headergw-srcIP, gw-destIP,
TTL, pktID, …
len
application datalenTCP headerlenInner IP header
host-srcIP, host-destIP,TTL, pktID, …
AHSPI,seqnum,
MAC, …
Inner AH MAC scope
Outer IP headergw-srcIP, gw-destIP,
TTL, pktID, …
len ESP
trailer: pad,padlen…
Outer ESP MAC scope
ESPheader:
SPI,seqnum
encryption scope
AH in transport modeHost-to-host auth-only,
e.g. network management?
ESP (conf and auth) in transport mode
Host-to-host secure channel,
e.g. encapsulated legacy apps?
ESP auth-only in transport mode
As for AH-transport; probably worse in all
cases…AH in tunnel modeBetween-gateway VPN
with no overall confidentiality provision
(may be good performance choice)ESP in tunnel modecarrying AH in transport
modeCommon “road warrior”
VPN: secure channel across public network, and strong, continuing authentication to end
system(s)
ESPauth: MAC
ESPauth: MAC
But where did the keys come from?
• “SPI” (security parameters index) refers to an “SA” (security association)
• SAs set up manually or by IKE – IPSec Key Exchange
• Policy “databases” define how fine- or coarse-grained SAs are– anything from “all traffic shall use this key” to
individual combinations of source and destination addresses and ports
– even “user-based” keying supported… but binding a user to an IP address is (very) problematic…
Outbound processing
• Lookup policy for this datagram– drop, pass through, or process
• Create a new SA if none exists• Apply keys from SA for MAC and enciphering• Add explicit IV for each datagram
– because they can be lost and arrive out-of-order
• Pass assembled datagram down to link layer– or to next instance of IPSec processing!– Let’s ignore fragmentation, PMTU discovery, …
Inbound processing
• Lookup policy for this datagram– drop, pass through, or process
• SA should already exist (we’re the responder)
• Apply keys from SA for MAC-check and deciphering (using datagram’s IV too)
• Raise security error if needed; otherwise,• Pass assembled datagram up to rest of
normal IP processing– or to next instance of IPSec processing!
What can be MAC’ed?• Immutable or predictable fields and options in IPv4 and IPv6• AH-controlled all immutable, so MACed:
– payload, payload length, next-header, SPI, sequence number, reserved fields
• IP header immutables and predictables v6:– version, payload length, next-header, source and destination IP
addrs,– BUT NOT: class, flow label, or hop limit (= v4 TTL)– all v6 extension headers self-describing as to mutability
• IP header immutables and predictables v4:– version, header and payload lengths, packetID, protocol type,
source and destination IP addrs– BUT NOT: ToS, flags, fragment offset, TTL, header checksum
• All unpredictable fields zero-filled for MAC calculation– so overall length and alignment still protected
IPsec Key Exchange – IKEor, time for a deep breath!
• Documentation hard to follow– IKE is specific adaptation of more general
protocols (“Oakley” and “ISAKMP”)
• Two levels of SA negotiated– an initial context (bidirectional, with heavy-
duty authentication and negotiation)– then several client SAs, negotiated quickly
using initial SA as secure channel; one for each direction and each AH and ESP
– initial SA also used for error traffic and similar management traffic
IKE security goals
• Authentication of parties (by digital signature, proof of knowledge of private key, or shared key)
• Establishment of a fresh shared secret• Shared secret used to derive keys for channel
confidentiality and authentication• “Perfect Forward Secrecy”, at cost of using up shared
material• (partial) anti-clogging, against denial-of-service
attacks• Secure negotiation of algorithms: asymmetric (e.g.
RSA, elliptic curve), symmetric (e.g. 3DES, Blowfish, AES), and hash (e.g. MD5, SHA-1)
IKE details
• Phase 1 is the heavyweight exchange to establish a secure key management channel– “Main mode” variant: slower, more cautious,
hides details of credentials used and allows forward secrecy (independence of short-term keys)
– “Aggressive mode”: less negotiation, fewer round trips, more information disclosed
• Phase 2 (“quick mode”) established SAs for IPSec itself, using the Phase 1 channel
Main Mode IKE Phase 1using digital signatures
(see RFC2409, p.11)
Initiator Responder
HDR, SA_i -->
<-- HDR, SA_rHeaders (HDR) include “cookies” CKY-I and CKY-R respectively. Initiator’s SA has one or more “proposals”, in preference order, for algorithms to be used for ISAKMP, the key management channel we’re building. The responder chooses exactly one of these proposals. These SAs may refer to one of a few standard Diffie-Hellman groups (both integer and ECC), or may define new DH groups.
HDR, KE, Ni -->
<-- HDR, KE, Nr
Ni, Nr are initiator’s and responder’s nonces, respectively; KE are Diffie-Hellman g^x and g^y valuesHDR*{IDii, [CERT,] SIG_I} -->
<-- HDR*{IDir, [CERT,] SIG_R}
HDR* denotes remaining ISAKMP traffic is encrypted. IDii, IDir are IDentifiers, typically IP addresses.SIG_I is over HASH_I = prf( SKEYID, g^x | g^y | CKY-I | CKY-R | SA_i | IDii )SIG_R is over HASH_R = prf(SKEYID, g^y | g^x | CKY-R | CKY-I | SA_i | IDir )where SKEYID = prf( Ni_b | Nr_b, g^xy ), and prf is the negotiated keyed pseudo-random function
Note 3 round-trips, 4 DH modular exponentiations, 2 signature-generations and 2 signature-verifications
Aggressive Mode IKE Phase 1using public-key proof
(see RFC2409, p.14)
HDR, SA_i, <Ni>Pubkey_r, <KE>Ke_i, <IDii>Ke_i -->
HDR includes CKY-I. SA_i has exactly one “take-it-or-leave-it” proposal. The nonce Ni is encrypted in the responder’s public key; KE (that’s g^x) and IDII are encrypted under Ke_i = prf( Ni, CKY-I ). So, the responder can decrypt Ni and so derive Ke_i only if it has the private complement to Pubkey_r.
<-- HDR, SA_r, <Nr>Pubkey_i, <KE>Ke_r,<IDir>Ke_r, HASH_R
HDR includes CKY-R. SA_r must equal SA_i. Similarly to the initiator’s message, the nonce Nr is encrypted in the initiator’s public key, while KE (that’s g^y) and IDir are encrypted under KE_r = prf( Nr, CKY-R ), requiring the initiator to have the private complement of Pubkey_i. HASH_R is as on the previous page.
HDR, HASH_I -->
The hashes sent in each direction aren’t signed; but the ability to generate them proves receipt and successful decryption of the nonce received from the other party.
Note 1.5 round-trips only, still 4 DH modular exponentiations, 2 public-key encrypts and 2 public-key decrypts. Careful analysis shows “plausible deniability”: the absence of digital signatures allows either party to disown the exchange.
Use of Phase 1 agreed material
Key material for the underlying ISAKMP key-management SA we’re building first is derived from the shared-secret quantity g^xy and the nonces securely exchanged during Phase 1 as follows:
SKEYID = prf( Ni | Nr, g^xy )
SKEYID_d = prf( SKEYID, g^xy | CKY-I | CKY-R | “0” )
SKEYID_a = prf( SKEYID, SKEYID_d | g^xy | CKY-I | CKY-R | “1” )
SKEYID_e = prf( SKEYID, SKEYID_d | g^xy | CKY-I | CKY-R | “2” )
where _a refers to Authenticator (MAC) material for the ISAKMP channel, and _e is for Encrypting material for the ISAKMP channel. _d is dual-purpose; firstly, it’s used as input for the _a and _e pseudo-random streams; secondly, it’s the main source of key material for the Phase 2 SAs which are the ones used by IPSec itself. SKEYID is used directly as the prf key for HASH_I and HASH_R, used to authenticate the parties.
Particular “transforms” (symmetric encryption algorithms, MACs, and so on) specify exactly how SKEYID_a, SKEYID_e, and SKEYID_d is to be used. For example, the specification for single-key DES uses at minimum the first 8 bytes of the PRF, forcing the parity bits to appropriate values, throwing away any bytes which would give rise to the known weak or semi-weak keys. (There are only 16 out of 2^56 such keys, so this isn’t likely to occur in practice!) The Triple-DES definition uses at least 24 bytes of the prf output, and the prf definition “stretches” its initial result by repeated application to produce as many bytes as are needed.
Phase 2: “Quick Mode”Now that we have an ISAKMP SA to define a secure key-management channel, doing algorithm and key agreement for client SAs such as AH and ESPis cheap and easy (relatively speaking). As example, this is how you get 4 SAs (one for each direction of an AH + ESP pair) - see RFC2409 p.19:
HDR*{HASH(1), SA0, SA1, Ni} -->
As before, HDR*{} means that all further material is encrypted (under SKEYID_e, remember?). SA0, SA1, etc. are “proposals” for client SAs for the AH and ESP transforms – each one is a preference-ordered list of possible algorithm combinations. Ni is a new initiator nonce. HASH(1) = prf( SKEYID_a, M-ID | SA0 | SA1 | Ni ); see how SKEYID_a is the MAC key. M-ID is the unique message-ID from HDR.
<-- HDR*{HASH(2), SA0, SA1, Nr}
Back come single algorithm choices for each SA, and a new responder nonce Nr. HASH(2) is similar to HASH(1): HASH(2) = prf( SKEYID_a, M-ID | Ni | SA0 | SA1 | Nr ); it has Ni added as a liveness proof.
HDR*{HASH(3} -->
This is a simple acknowledgement that the responder’s message has been received;HASH(3) = prf(SKEYID_a, “0” | M-ID | Ni | Nr )
Now key material for each IPSec SA is defined as follows:
KEYMAT = prf( SKEYID_d, protocol | SPI | Ni | Nr )
Since protocol and SPI are unique to AH/ESP and direction respectively, this gives 4 separate chunks of KEYMAT. If necessary, they are “stretched” as before by applying prf iteratively. Note the single Phase 1 DH exchange and public-key operations have been used to derive key material for all four IPSec SAs, spreading the cost of those expensive operations. There’s an option to include a fresh DH exchange in each Quick Mode if you prefer Forward Secrecy to computational efficiency...
Final notes on IPSec
• IKE is carried over UDP; hence unreliable (may need to be restarted) and blocked by some firewalls
• Managing IPSec policy and deployments isn’t easy, and getting it wrong can be embarassing in losing connectivity, e.g. by making exchanges of routing updates unreadable
• After trying to roll-its-own with PPTP, MS has put IPSec into WinXP
• See FreeS/WAN for implementation (and contribute too, unless you’re a US citizen):http://www.xs4all.nl/~freeswan/