measurement and analysis of private key sharing in the...

Measurement and Analysis of Private Key Sharing in

the HTTPS Ecosystem

Frank Cangialosi, Taejoong Chung, David Choffnes, Dave Levin, Bruce M. Maggs, Alan Mislove, Christo Wilson

How do we know with whom we are communicating?

Certificate

Public

Private

Certificate

CertificateAuthorities

Public

Private

Certificate

Certificate


CertificateCertificate



TLS Handshake



Authentication fundamentally assumes:

Only knows


The PKI in today’s web

Rare!


Third-party Hosting Providers

• Content delivery networks

• Web hosting services

• Cloud providers

Varying levels of involvement

But all trusted to deliver content

Certificate


Third-party hosting providers know their customers’ private keys

Third-party hosting providers know their customers’ private keys

Authentication fundamentally assumes:Only knows

Example of key sharing

What’s wrong with sharing?

1. Complicates the trust model,users don’t know who they’re really trusting

2. Potential to create centralization of trust

3. Potential to create single point of failure (in terms of management)

This study

This study

How many websites share their private keys?

This study

How many keys have 3rd parties obtained?


This study

How has this affected key management?

How many keys have 3rd parties obtained?


How do we detect sharing at scale?Rapid7 weekly port 443 scans 2013-2015DATA

DomainDomainDomainDomainCertificateIP Addr



IPv4 Scan DomainDomainDomainDomainCertificateIP Addr



DomainDomainDomainDomainCertificate


IP Addr


IPv4 Scan



IP Addr




IP Addr

How do we detect sharing at scale?


DomainDomainDomainDomainCertificate IP Addr

IP Addr

Rapid7 weekly port 443 scans 2013-2015DATA



DomainDomainDomainDomainCertificate IP Addr

IP Addr

5.1 million valid leaf certificatesRapid7 weekly port 443 scans 2013-2015DATA


Domain

Domain

Domain

Domain

Domain

IP Addr

IP Addr

5.1 million valid leaf certificates


Domain

Domain

Domain

Domain

Domain

IP Addr

IP Addr

Does the same entity that owns the domain own and operate the server at that IP address?


Domain

Domain

Domain

Domain

Domain

IP Addr

IP Addr

Reverse DNS


Domain equivalence?

google.com google.co.uk

google.com youtube.com

nestle.com friskies.com

whitehouse.gov whitehouse.com

Domain names alone are not enough

http://google.com

http://google.co.uk

http://google.com

http://youtube.com

http://nestle.com

http://friskies.com

http://whitehouse.gov

http://whitehouse.com

Incorporating whois

Emails in whois records reflect administrative domain

google.com

google.co.uk

google.de

zagat.com

golang.org

Incorporating whois


google.com

google.co.uk

google.de

zagat.com

golang.org

whois RegistrantEmail:

AdminEmail:

TechEmail:

[email protected]

[email protected]

[email protected]

Incorporating whois


google.com

google.co.uk

google.de

zagat.com

golang.org

[email protected]@[email protected]

Incorporating whois


google.com

google.co.uk

google.de

zagat.com

golang.org

RegistrantEmail:

AdminEmail:

TechEmail:

[email protected]

[email protected]

[email protected]

whois

whois


Incorporating whois


google.com

google.co.uk

google.de

zagat.com

golang.org



Incorporating whois

google.com

google.co.uk

google.de

zagat.com

golang.org


whois

whois


Incorporating whois

google.com

google.co.uk

google.de

zagat.com

golang.org


RegistrantEmail:

AdminEmail:

TechEmail:

[email protected]

[email protected]

[email protected]

whois

whois


Incorporating whois

google.com

google.co.uk

google.de

zagat.com

golang.org


Incorporating whois


google.com

google.co.uk

google.de

zagat.com

golang.org

Incorporating whois

Domain Organization



Domain

Domain

Domain

Domain

Domain

IP Addr

IP Addr


Domain

Domain

Domain


Domain

Domain Org

Domain Org

DomainDomain Org IP Addr

IP Addr


Domain

Domain

Domain


Domain

Domain Org

Domain Org

DomainDomain Org Host

Host


Key sharing: domain org ≠ host org

Domain

Domain

Domain


Domain

Domain Org

Domain Org


HostHostOrg

HostOrg


Key sharing: domain org ≠ host org

Outline

How many keys have providers aggregated?

How does sharing impact key management?

How prevalent is key sharing?


Domain

Domain

Domain

Domain

Domain Org

Domain Org


HostHostOrg

HostOrg


Domain

Domain

Domain

Domain

Domain Org

Domain Org


HostHostOrg

HostOrg

HostOrg

Domain Org


Domain

Domain

Domain

Domain

Domain Org

Domain Org


HostHostOrg

HostOrg

HostOrg

Domain Org

Self-hosted

Key sharing


0

0.2

0.4

0.6

0.8

1

0 1 10 102 103 104 105

CD

F

Number of Third-Party Hosting Providers Used

Organizations


0

0.2

0.4

0.6

0.8

1

0 1 10 102 103 104 105

CD

F


Organizations23.5% Self-hosted


0

0.2

0.4

0.6

0.8

1

0 1 10 102 103 104 105

CD

F



76.5%share

at least1 key


0

0.2

0.4

0.6

0.8

1

0 1 10 102 103 104 105

CD

F



76.5%share

at least1 key

Who?

Who shares?

0

0.2

0.4

0.6

0.8

1

0 200k 400k 600k 800k 1M

Fra

cti

on

of

Do

ma

ins

Ho

ste

do

n T

hir

d-p

art

y P

rov

ide

rs

Alexa Site Rank (bins of 10,000)

At least one key sharedAll keys shared

Who shares?

0

0.2

0.4

0.6

0.8

1

0 200k 400k 600k 800k 1M

Fra

cti

on

of

Do

ma

ins

Ho

ste

do

n T

hir

d-p

art

y P

rov

ide

rs


At least one key sharedAll keys shared43.2% (of Top 10k)

share at least one

Who shares?

0

0.2

0.4

0.6

0.8

1

0 200k 400k 600k 800k 1M

Fra

cti

on

of

Do

ma

ins

Ho

ste

do

n T

hir

d-p

art

y P

rov

ide

rs



share at least one

22.4% share all

Who shares?

Key sharing is common across the Internet

0

0.2

0.4

0.6

0.8

1

0 200k 400k 600k 800k 1M

Fra

cti

on

of

Do

ma

ins

Ho

ste

do

n T

hir

d-p

art

y P

rov

ide

rs



share at least one

22.4% share all

Outline




Outline




• 76.5% share with ≥ 1 provider • Common even among most popular websites

Domain

Domain

Domain

Domain

Domain Org

Domain Org


HostHostOrg

HostOrg

HostOrg

Domain Org


100

101

102

103

104

105

106

100

101

102

103

104

105

106

Nu

mb

er

of

Dis

tin

ct

Cu

sto

mers

Serv

ed

Rank-Order Third-Party Hosting Providers

How have keys been aggregated?

100

101

102

103

104

105

106

100

101

102

103

104

105

106

Nu

mb

er

of

Dis

tin

ct

Cu

sto

mers

Serv

ed



secureserver.net

unifiedlayer.comamazonaws.com

Cloud Flare Inc.Rackspace Hosting

akamaitechnologies.com

266,110151,628117.22978,36954,158

15,440… …

#Organizations Hosting provider277,891175,089122,158

87,07763,418

22,671…

#Domains

100

101

102

103

104

105

106

100

101

102

103

104

105

106

Nu

mb

er

of

Dis

tin

ct

Cu

sto

mers

Serv

ed



secureserver.net

unifiedlayer.comamazonaws.com

Cloud Flare Inc.Rackspace Hosting

akamaitechnologies.com

266,110151,628117.22978,36954,158

15,440… …

#Organizations Hosting provider277,891175,089122,158

87,07763,418

22,671…

#Domains

Top 1% of providers hold keysfor 86% of all organizations

Domain

Domain

Domain

Domain

Domain Org

Domain Org


HostHostOrg

HostOrg

HostOrg

Domain Org

Does key sharing make enticing attack targets?


0

0.2

0.4

0.6

0.8

1

100 101 102 103 104 105 106

Cu

mu

lati

ve

Fra

cti

on

of

Do

ma

ins

’ K

ey

s A

cq

uir

ed

Number of Hosting Providers Compromised

Alexa Top 1kAlexa Top 1m

All Domains


0

0.2

0.4

0.6

0.8

1

100 101 102 103 104 105 106

Cu

mu

lati

ve

Fra

cti

on

of

Do

ma

ins

’ K

ey

s A

cq

uir

ed



All Domains

60% of Top 1K, same provider


0

0.2

0.4

0.6

0.8

1

100 101 102 103 104 105 106

Cu

mu

lati

ve

Fra

cti

on

of

Do

ma

ins

’ K

ey

s A

cq

uir

ed



All Domains

>40% of all sites, 10 providers



0

0.2

0.4

0.6

0.8

1

100 101 102 103 104 105 106

Cu

mu

lati

ve

Fra

cti

on

of

Do

ma

ins

’ K

ey

s A

cq

uir

ed



All Domains

Popular hosting services are prime targets for attack

>40% of all sites, 10 providers


Outline




• Top 1% of providers hold keys for 86% of orgs

• Attractive targets for attack


Key Management

Request certificates

Renew expiring certificates

Revoke and reissue compromised certificates

Who manages private keys?

CAsWebsite acquires Third-party acquires


Website acquires Third-party acquires



Diverse

“Self-managed”



Diverse Heavily skewed

“Self-managed” “Outsourced”



Diverse Heavily skewed

“Self-managed” “Outsourced”58.4% of Alexa Top 10K 33.0% of all domains


0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date

Self-managedOutsourced

Natural experiment: Heartbleed (4/7/2014)


0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


CloudFlarerevocations



0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


Outsourced (w/o CF)




0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


Outsourced (w/o CF)


Slightly more thorough



0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


Outsourced (w/o CF)


10 days to react!




0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


0.75

0.8

0.85

0.9

0.95

1

04/07 04/11 04/15 04/19 04/23 04/27 05/01 05/05

Fra

cti

on

of

Cert

ific

ate

s

No

t R

evo

ked

Date


Outsourced (w/o CF)


10 days to react!



A few revoked thoroughly, but many did not!


0

0.2

0.4

0.6

0.8

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

CD

F o

f H

os

tin

g P

rov

ide

rs

Fraction of Heartbleed-vulnerable Certificates Revoked



(One year after Heartbleed)


0

0.2

0.4

0.6

0.8

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

CD

F o

f H

os

tin

g P

rov

ide

rs



66% of providers did not revoke a single vulnerable certificate!




0

0.2

0.4

0.6

0.8

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

CD

F o

f H

os

tin

g P

rov

ide

rs



66% of providers did not revoke a single vulnerable certificate!

A small minority of providers revoked most of their vulnerable certificates, but none revoked all



Outline




• Creates single point of failure • Most third-parties did poor job of revoking

• Top 1% of providers hold keys for 86% of orgs

• Attractive targets for attack


Due to economic incentives, key sharing is prevalent in today’s web

Future work on the PKI should take economics and hosting providers into account, ideally:

hosting should not require key sharing

Most providers are not managing keys responsibly

[email protected]

securepki.orgCode and data available at:

measurement and analysis of private key sharing in the...

Documents