• 沒有找到結果。

The Internet is Too Secure Already


Academic year: 2022

Share "The Internet is Too Secure Already"


加載中.... (立即查看全文)



USENIX Security 2003 1

The Internet is Too Secure Already

Eric Rescorla RTFM, Inc.


USENIX Security 2003 2

Overview of my argument

We have lots of communications security tech.


But actual Internet communications are insecure


We have the wrong threat model!

We worry about all known threats

Too good security is trumping deployment Practical security isn’ t glamorous


USENIX Security 2003 3

Structure of this talk

Overview of the current situation

Internet threat model

Real protocol deployment

Where is the effort going and why?

How could we make things better?

Appropriate threat models Better customer models


USENIX Security 2003 4

Structure of this talk

Overview of the current situation

Internet threat model

Real protocol deployment

Where is the effort going and why?

How could we make things better?

Appropriate threat models Understand our customers


USENIX Security 2003 5

The Internet threat model

Attacker has complete control of the network

Can modify, delete, insert, duplicate, etc.

“ Hand packets to attacker to deliver”

End-systems more or less inviolate

Not really true

… but hard to do communications security without it

Don’ t get embarassed


USENIX Security 2003 6

Real attacks are less glamorous

Remote penetration

Find simple programming bugs

Buffer overflows Format strings…

Mostly a matter of effort…


Viruses Worms



USENIX Security 2003 7

Structure of this talk

Overview of the current situation

Internet threat model

Real protocol deployment

Where is the effort going and why?

How could we make things better?

Appropriate threat models Understand our customers


USENIX Security 2003 8

Two wins, Three draws, One loss

Before we can analyze we need data Two wins


Three draws


One loss



USENIX Security 2003 9

SSL/TLS Status

Main protocol for Web security

And other kinds of channels

Quite mature

SSLv2 released in 1994

Not very good

SSLv3 released in 1995 TLSv1 published in 1999 Now working on TLS 1.1

TLS 1.1?

General cleanup + fix for Rogaway attack


USENIX Security 2003 10

SSL/TLS Deployment

Very widely implemented

Nearly all browsers/servers have it Lots of Open Source toolkits



Tens of billions of dollars in transactions

… but only about 1% of servers

And most of them have invalid certs


Sporadic usage for SMTP, IM, etc.

Certificates almost always self-signed

Most common use of TLS is for e-commerce

but credit card liability is limited Free rider problem


USENIX Security 2003 11

Why has SSL succeeded?

It’ s easy to use

Interface looks almost exactly like what it replaces Just type “ https” instead of “ http”

Can be deployed without much external help

Certificates are relatively easy to get

And only servers need them

Especially if you let mod_SSL make you a “ Snake Oil” one

There is a real incentive

Credit card sniffing was scary

and there was big money to be made


USENIX Security 2003 12

SSH Status

Premier secure remote login protocol

Originally invented by Tatu Ylonen in 1994 Program was the spec

Security status

Lots of holes in SSHv1 SSHv2 pretty good

Standardization has really lagged

IETF standard version due out Real Soon Now But the protocol is pretty mature now


USENIX Security 2003 13

SSH Deployment

Near-universal on Unix

Available on Ciscos, etc.

Clients available for Windows, Mac, Java

Arguably the most successful security protocol

Less total use than SSL

But completely dominates its market segment

Telnet and rsh have essentially vanished


USENIX Security 2003 14

Why has SSH succeeded?

It’ s easy to use

Interface looks almost exactly like what it replaces

alias ‘ rsh’ to ‘ ssh’

Can be deployed without any external help

Both parties to the transaction know each other Leap of faith authentication

There is a real incentive

Password sniffing was a real problem Sysadmins can impose it on users VPNs are a pain


USENIX Security 2003 15

SSH’ s Leap of Faith

Problem: client needs server’ s public key

Don’ t want to use certificates


Server gives client bare public key

MITM possible

Optional verification with fingerprints

Client caches server’ s key

Detects changes

This was not well received originally

But now it’ s considered clever


USENIX Security 2003 16

IPsec Status

Security for IP traffic Two main pieces

Packet formats

Authentication Header (AH)

Encapsulating Security Payload (ESP)

Key management: IKE (Internet Key Exchange)

AH/ESP basically mature

Though getting tweaked

IKE getting a total rewrite

This is very late.

Planned for 2001 Due Real Soon Now


USENIX Security 2003 17

IPsec Deployment

Widely implemented

Built into Windows, Solaris, Cisco

Available for Linux (FreeS/WAN), FreeBSD (KAME)

Only really used for VPNs

Using dedicated appliances Manual configured

Shared static keys Self-signed certs

Being replaced by SSL VPN!


USENIX Security 2003 18

S/MIME Status

One of two primary e-mail encryption protocols

Designed by RSA

S/MIME v2 stable and mature

S/MIME v3 currently under development Minor tweaks only…

Replacing DH/DSS with RSA

Reversal of previous patent evasion

X.400 gatewaying

Symmetric key distribution


USENIX Security 2003 19

S/MIME Deployment

In a number of major mail programs

Outlook, OE, Netscape

Almost totally unused

PGP has more users

But not many

Really hard to get certificates

Where can I get my own cert?


How hard is it?

It takes hours!

And what does it promise?

E-mail validity

I waited for that????

Other people’ s certs are in hiding


USENIX Security 2003 20

Do people just not want secure e-mail?

This is our third run at the wall!

At least…

PEM, MOSS are direct ancestors

Also PGP, DMS, X.400, OpenPGP

Nobody wanted any of the others either But people say they want secure e-mail

And VCs believe it…

Voltage, PGP Inc., SIGABA, Tumbleweed

So what’ s the story?


USENIX Security 2003 21

PKIX Status

IETF Standard for certificates 8 years old

Lots of output

18 RFCs

1.5 MB total

And still plugging away

28 I-Ds

1.7 MB total

Plus, PKI Forum…

Will we ever be done?


USENIX Security 2003 22

PKIX Deployment

Lots of implementations

You get a CA for free with Windows Advanced Server!

But interoperability is a nightmare

Unless you stick to the common subset

Internet deployment limited to SSL

And self-signed certs are common

Enterprises bought PKI

But it made them miserable

… and they don’ t deploy it


USENIX Security 2003 23

The WEP Debacle

“ Security” for 802.11 WEP is badly broken The big problem

The channel security misused RC4

Most common crypto error ever

Tools exist to break into any WEP network in minutes

The small problem

Key management is simple shared key

Probably not the best idea

These problems are being fixed

Still waiting for a final standard (TKIP, 802.11i) Current deployed systems are broken


USENIX Security 2003 24

WEP Deployment Status

In almost every 802.11 card and AP Not always turned on

28 % of networks use it

And those networks are easily crackable

People seem to be scared by the publicity

Still a lot more deployment than IPsec

And a heck of a lot better than nothing


USENIX Security 2003 25

Common themes

Use lags availability

Just having the stuff there isn’ t enough

Certificates are really hard to get

Blocker for S/MIME, IPsec Partial blocker for SSL

Wide use of weak certificates

This stuff is too hard to use

See “ Why Johnny Can’ t Encrypt”

Do usage model first

Then get security right (SSL, SSH)


USENIX Security 2003 26

Some possible explanations

Security is inherently hard to use

Possible but doesn’ t get us anywhere

The customer is stupid

Probably true, but he’ s not getting any smarter

We’ re delivering the wrong products

We’ ll sell no wine before its time

But we’ve been working on this stuff for > 10 years

We’ re using the wrong design criteria

So the end product is undesirable

This is the only theory that gets us somewhere


USENIX Security 2003 27

The wrong design criteria?

This causes two kinds of problems

Intentionally building the wrong product

Because we think it’ s the right one (IPsec)

Diverting resources due to feature misprioritization

Emphasis on security over usability (Name-based virtual hosts)

Criteria cannot be derived from first principles

You have to know the customer


USENIX Security 2003 28

Structure of this talk

Overview of the current situation

Internet threat model

Real protocol deployment

Where is the effort going and why?

How could we make things better?

Appropriate threat models Understand our customers


USENIX Security 2003 29

Where is the effort going?

Inventing new mechanisms


Stream authentication New cipher modes

Polishing existing protocols

Defenses against impractical attacks New security features

Replacing old algorithms with new ones


The occasional actual improvement


USENIX Security 2003 30

Name-based virtual hosting

HTTP virtual servers

Multiple web servers on a single physical server Disambiguated by the Host: header

But don’ t work with HTTPS

Need to know virtual host to choose certificate But SSL handshake happens first

So you don’t see the Host: header till too late

With SSL you need 1 IP address per virtual host

Fix: put the name in the SSL handshake

Done in Domain name extension But held hostage to..

Packet size, external certs, OCSP…


USENIX Security 2003 31

Current work on SSL/TLS: Attacks

Kocher/Boneh/Brumley timing attack

Extract a private key But how practical is it?

several million trials on an intranet Billions on a WAN?

OpenSSL finally fixes it…

Vaudenay CBC attack

Extract passwords from automated clients

Rogaway CBC attack

Verify a guess of a single cipher block

Bad Version Oracles

Recover a single session key

Extension to Bleichenbacher’s attack

Requires a million trials


USENIX Security 2003 32

Rogaway CBC Attack

Attacker can verify guess of ciphertext

By injecting a chosen plaintext And observing

This only works well when SSL is used in a proxy

SSL Engine

Plaintext Ciphertext


Inject here Read here


USENIX Security 2003 33

Current work on SSL/TLS: Responses to Attacks

Wide publicity

Most of these attacks lead to papers

Vaudenay, Kocher/Boneh/Brumley attacks got coverage

Immediate fixes issued

To OpenSSL

New version of TLS

No known actual attacks in wild

No known available tools

Contrast with OpenSSL buffer overflows

Slapper released within 2 months


USENIX Security 2003 34

Current work on IPsec

AH/ESP are basically unchanged IKE being totally redone (IKEv2)

This has taken 2 years!

What issues are holding us up?

Cipher suites vs. a la carte Identity protection?

6 messages or 4 Provable exchange

Did I mention it still doesn’ t work?

Certificates and fragmentation

I am not making this up!


USENIX Security 2003 35

Why do IKEv2 anyway?

Nobody wants IKEv1

Complaints that it’ s too hard to implement

Vague specification

Extremely complex protocol

… but there are lots of interoperable implementations

VPNC lists >10 conformant implementations

The real reason?

We’ re flailing

Nobody uses IKE

so we have to try something


USENIX Security 2003 36

What’ s the story with S/MIME?

The protocol is in good shape Everyone has it, noone uses it The problem is certificates


But noone has them


USENIX Security 2003 37

PKIX, the standard that won’ t die

Hideously complex

RFC 3280 is 129 pages long Lawyers are involved!

Noone knows what anything means


Comparison Structure

keyUsage (nonRepudiation) Constraints


And they don’ t implement it anyway

CRL checking

Constraints again

All I wanted was to authenticate who I was talking to!


USENIX Security 2003 38

Structure of this talk

Overview of the current situation

Internet threat model

Real protocol deployment

Where is the effort going and why?

How could we make things better?

Appropriate threat models Understand our customers


USENIX Security 2003 39

Three examples of threat model mismatch

Excessive concern with active attacks

The easiest attacks are passive

Leads to requirement for certificates

Taking cryptanalytic attacks too seriously

Leads to protocol churn

Not bad in itself, but very distracting

Forgetting about other threats

User stupidity Software holes


USENIX Security 2003 40

Why isn’ t there any tooling to steal private keys?

Kocher’ s timing attack is years old

But no tools are available We know it’ s possible now…

But still no tooling

Maybe it’ s still too hard?

The OpenSSL exploits didn’ t steal private keys

This would have been incredibly easy

The keys are generally just in memory

Other exploits don’ t seem to either

When the Web server is compromised it’ s easy

Maybe people don’ t want them…


USENIX Security 2003 41

Maybe private keys aren’ t so important?

Using a stolen key is harder than it looks

Pretty much requires being on the same network as victim

People’ s information isn’ t that interesting

Credit card numbers are easy to get

Buy on the black market

Break into e-commerce servers

SSH keys are only useful for breakins

But once you’ ve already broken in…


USENIX Security 2003 42

The worst case happens.. And noone notices!

What happens if a CA is compromised?

An attacker can impersonate everyone

Pretty bad, huh?

IE cert verification was totally broken until 2002

Basic constraints verification broken

This means that anyone can forge certificates!

This is worse than a CA compromise

Since it can’t be fixed with CRLs

Lots of people still have broken versions

Because they haven’t upgraded

And yet no rash of attacks


USENIX Security 2003 43

User Stupidity (I): SSL Certificates

Clicking “ Yes” may not be your best plan here!


USENIX Security 2003 44

User Stupidity (II): Executable email

Windows allows executable email

VBScript Javascript

Actual Windows binaries

Users are asked before .EXEs are run

And they often say yes

Worms often spread this way

How can secure e-mail work in this environment?


USENIX Security 2003 45

Bugs in software

Holes found in most COMSEC implementations

Buffer overflows



Failure to correctly perform protocols


All of these bugs were worse than 99% of protocol failures

“ All software has bugs. Security software has security-relevant bugs”

-- Steve Bellovin


USENIX Security 2003 46

What’ s an appropriate threat model?

Worry a lot about passive attack

Else why bother at all?

Worry about active attack

But not if it means making things undeployable Lesson of SSH

Leap of faith

Don’ t worry at all about being embarrassed

Unless you did something really stupid


USENIX Security 2003 47

Structure of this talk

Overview of the current situation

Internet threat model

Real protocol deployment

Where is the effort going and why?

How could we make things better?

Appropriate threat models Understand our customers


USENIX Security 2003 48

Customers lie

But not all the time

Our job is to know what they want

and to try to give it to them or they won’ t take it...

What they want may not be what we think they

should want


USENIX Security 2003 49

“ Security is really important”

this means...


USENIX Security 2003 50

“ Security is really important”

this means...

“ The appearance of security is really important”

These are not the same thing

He wants to know what to tell his boss


USENIX Security 2003 51

“ Security is more important than features”

this means...


USENIX Security 2003 52

“ Security is more important than features”

this means...

“ I want my dancing pigs”

In the battle between features and security features always win

Active content, firewall bypassing, Windows...

Don’ t torture your users


USENIX Security 2003 53

“ Make security easy to install”

this means...


USENIX Security 2003 54

“ Make security easy to install”

this means...

“ It better just drop in and work”

Need I say more?


USENIX Security 2003 55

An Agenda: Evidence-based Security

General problem: what security measures make a difference

What threats are most serious?

Which ones can we fix

And at what cost?

What will users deploy?

These questions can’ t be answered a priori

It requires unglamorous research


USENIX Security 2003 56

What threats are most serious?

This is probably where we have the most data

Market research

… but it’s spotty

Companies tend to hide this information

Not too much academic research

My impressions

Cryptanalytic attacks are really rare Protocol flaws are rare

Attacks on programming flaws are common DDoS is common


USENIX Security 2003 57

Which threats are easy to fix?

Some gut reactions

Our protocols are about as secure as they’ re going to get

But we can make them more deployable

But we can fix our code problems

Stop using C

Sandboxing, compiled in protection

Janus, Systrace, *guard

Code checking tools

Metacompilation, RATS, Lint…

We don’ t know how to really fix DDoS

This is going to take some measurement

User/Programmer experience System performance

Effect on bug rate


USENIX Security 2003 58

What will the customers buy?

Rule of thumb: if we’ ve spent a lot of time on it and noone wants it, something is wrong

IPsec, X.509, secure e-mail

Painting it a different color probably won’ t help

Compliance is key

Side effects

Perceived cost/benefit ratio

Users may not make the choices we would


USENIX Security 2003 59

Questions we need to answer

What’ s the total cost of exposure of various kinds of threats?

How much are people willing to pay for various security features?

Why can't users use cryptographic protocols?

What percentage of security protocol features see implementation?

What sorts of implementation errors are most serious?

What programming practices would minimize them?

What's the cost of upgrades?

What's the cost of obtaining information about vulnerabilities?

What sort of incentives would cause users to keep up to date?



Management Committees/ School Management Committees of their sponsored schools (hereafter collectively referred to as “school governance authority”) in formulating specific

Courtesy: Ned Wright’s Cosmology Page Burles, Nolette & Turner, 1999?. Total Mass Density

You need to configure Windows Firewall with Advanced Security on Server1 to allow the ping utility

Teacher / HR Data Payroll School email system Exam papers Exam Grades /.

Classifying sensitive data (personal data, mailbox, exam papers etc.) Managing file storage, backup and cloud services, IT Assets (keys) Security in IT Procurement and

The roles of school management and technical support staff on implementing information and network security measures... Security

 Service Level Agreement – ensure at least 99.7% availability of the WiFi service, support four-hour response time and four-hour service recovery with active monitoring,

• A delta-gamma hedge is a delta hedge that maintains zero portfolio gamma, or gamma neutrality. • To meet this extra condition, one more security needs to be