Share
Embed
Scalable TLS
Artyom "Töma" Gavrichenkov
November 07, 2016
Transcript
77 slides
-
Краткая история нового времени • 2010: SPDY w/de-facto mandatory* SSL/TLS
• 2014: “HTTPS as a ranking signal” at Google • 2015: HTTP/2 w/de-facto mandatory* TLS • 2016: Let’s Encrypt * – https://forum.nginx.org/read.php?21,236132,236184 *– https://daniel.haxx.se/blog/2015/03/06/tls-in-http2/ -
Краткая история нового времени • 2010: SPDY w/de-facto mandatory* SSL/TLS
• 2014: “HTTPS as a ranking signal” at Google • 2015: HTTP/2 w/de-facto mandatory* TLS • 2016: Let’s Encrypt * – https://forum.nginx.org/read.php?21,236132,236184 *– https://daniel.haxx.se/blog/2015/03/06/tls-in-http2/ -
Краткая история нового времени • 2010: SPDY w/de-facto mandatory* SSL/TLS
• 2013: NSA story • 2014: “HTTPS as a ranking signal” at Google • 2014: • 2015: HTTP/2 w/de-facto mandatory* TLS • 2015: • 2016: Let’s Encrypt • 2016: * – https://forum.nginx.org/read.php?21,236132,236184 *– https://daniel.haxx.se/blog/2015/03/06/tls-in-http2/ -
Краткая история нового времени • 2010: SPDY w/de-facto mandatory* SSL/TLS
• 2013: NSA story • 2014: “HTTPS as a ranking signal” at Google • 2014: Heartbleed, POODLE • 2015: HTTP/2 w/de-facto mandatory* TLS • 2015: RFC 7457 • 2016: Let’s Encrypt * – https://forum.nginx.org/read.php?21,236132,236184 *– https://daniel.haxx.se/blog/2015/03/06/tls-in-http2/ -
Краткая история нового времени • 2010: SPDY w/de-facto mandatory* SSL/TLS
• 2013: NSA story • 2014: “HTTPS as a ranking signal” at Google • 2014: Heartbleed, POODLE • 2015: HTTP/2 w/de-facto mandatory* TLS • 2015: RFC 7457 • 2016: Let’s Encrypt * – https://forum.nginx.org/read.php?21,236132,236184 *– https://daniel.haxx.se/blog/2015/03/06/tls-in-http2/ -
Краткая история нового времени • 2010: SPDY w/de-facto mandatory* SSL/TLS
• 2013: NSA story • 2014: “HTTPS as a ranking signal” at Google • 2014: Heartbleed, POODLE • 2015: HTTP/2 w/de-facto mandatory* TLS • 2015: RFC 7457, FREAK, Logjam • 2016: Let’s Encrypt • 2016: DROWN * – https://forum.nginx.org/read.php?21,236132,236184 *– https://daniel.haxx.se/blog/2015/03/06/tls-in-http2/ -
SSL/TLS PKI • Root certificate authorities, trust chain • 92
CAs in Firefox -
SSL/TLS PKI • Root certificate authorities, trust chain • Trusted,
because they make it for living • Independent from large corporations, government, etc. -
SSL/TLS PKI • Root certificate authorities, trust chain • Trusted,
because they make it for living • Independent from large corporations, government, etc. Except, some of them ARE government -
SSL/TLS PKI • Root certificate authorities, trust chain • Trusted,
because they make it for living • Independent from large corporations, government, etc. And some of them are large corporations Except, some of them ARE government -
SSL/TLS PKI • Root certificate authorities, trust chain • Trusted,
because they make it for living • Independent from large corporations, government, etc. • Pursuing their interests as trusted third parties -
SSL/TLS PKI • Root certificate authorities, trust chain • Trusted,
because they make it for living • Independent from large corporations, government, etc. • Pursuing their interests as trusted third parties • Corporations and government always tend to elevate their own interests -
The story of WoSign • Trusted since 2009 • Aggressive
marketing and free certificates • Passed audit by Ernst&Young -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control • Allowed using subdomains to verify 2nd level domain -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control • Allowed using subdomains to verify 2nd level domain • Allowed using arbitrary files to verify ownership -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control • Allowed using subdomains to verify 2nd level domain • Allowed using arbitrary files to verify ownership • Allowed to issue certificates for arbitrary domains without verification -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control • Allowed using subdomains to verify 2nd level domain • Allowed using arbitrary files to verify ownership • Allowed to issue certificates for arbitrary domains without verification • Issued backdated SHA-1 certificates -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control • Allowed using subdomains to verify 2nd level domain • Allowed using arbitrary files to verify ownership • Allowed to issue certificates for arbitrary domains without verification • Issued backdated SHA-1 certificates • Used unpatched software (such as dig) on the validation server -
The story of WoSign https://wiki.mozilla.org/CA:WoSign_Issues • Issued certificates not requested
by domain owner • Allowed using non-privileged ports (>50,000) to verify domain control • Allowed using subdomains to verify 2nd level domain • Allowed using arbitrary files to verify ownership • Allowed to issue certificates for arbitrary domains without verification • Issued backdated SHA-1 certificates • Used unpatched software (such as dig) on the validation server • Purchased other CA (StartCom) and attempted to suppress information about the ownership transfer -
The story of WoSign The aftermath? • Banned by Google
in Chrome • Banned by Mozilla for a year -
The story of WoSign The aftermath? • Banned by Google
in Chrome • Banned by Mozilla for a year • Still trusted by Microsoft and lots of unpatched equipment -
Aftermath • Go and choose the cheapest CA available •
Bonus points if it provides some kind of API -
Aftermath • Go and choose the cheapest CA available •
Bonus points if it provides some kind of API • Pick multiple CAs -
Aftermath • Go and choose the cheapest CA available •
Bonus points if it provides some kind of API • Pick multiple CAs • “Extended validity” certificates? -
Aftermath • Go and choose the cheapest CA available •
Bonus points if it provides some kind of API • Pick multiple CAs • “Extended validity” certificates are a security theater (don’t bother if you are not a bank and auditor doesn’t force you to) -
Aftermath • Go and choose the cheapest CA available •
Bonus points if it provides some kind of API • Pick multiple CAs • “Extended validity” certificates are a security theater (don’t bother if you are not a bank and auditor doesn’t force you to) • Prefer short-lived certificates -
Long-living certificates? Pros: • Discount • Less pain in the
#^$ updating all the certs -
Long-living certificates? Pros: • Discount • Less pain in the
#^$ updating all the certs Cons: • Soft-fail CRL and OCSP are not reliable • Hard-fail CRL and OCSP are never used (you may do it in your app though) • Certificate deployment and management must be automated anyway -
Long-living certificates? • CRL and OCSP are not reliable •
Certificate deployment and management must be automated Long-lived cert is a technical debt. It wouldn’t punish you immediately. It will hurt you eventually. -
Automated certificate management • Add, remove, change and revoke your
certificates real quick • Manage certificates properly: short lifetime, multiple keys • Set up a clientside TLS auth -
Automated certificate management • Add, remove, change and revoke your
certificates real quick • Manage certificates properly: short lifetime, multiple keys • Set up a clientside TLS auth • Quickly work around obscure issues like “Intermediate CA was revoked” -
The story of GlobalSign • During a planned maintenance, accidentally
revoked its own certificate • Used CDN (Cloudflare) for CRL and OCSP • Undid revocation, but it’s got cached on CDN -
The story of GlobalSign • During a planned maintenance, accidentally
revoked its own certificate • Used CDN (Cloudflare) for CRL and OCSP • Undid revocation, but it’s got cached on CDN • Four days before cached response will expire in a browser • Wikipedia, Dropbox, Spotify, Financial Times affected • Large sites affected more because CRL got cached everywhere immediately -
The story of GlobalSign • Large sites affected more because
CRL got cached everywhere immediately • “All is good and yet traffic dropped by 30%” • Really hard to troubleshoot • The issue is of distributed nature • You depend on a vendor -
The story of GlobalSign • Large sites affected more because
CRL got cached everywhere immediately • “All is good and yet traffic dropped by 30%” • Really hard to troubleshoot • The issue is of distributed nature • You depend on a vendor • Multiple different certs from different vendors helped to track down • tcpdump also of a great help: sessions got stuck at TLS Server Hello -
The story of GlobalSign • Really hard to troubleshoot •
The issue is of distributed nature • You depend on a vendor • Multiple different certs from different vendors will help to track down • tcpdump also of a great help: sessions got stuck at TLS Server Hello TLS is still bleeding edge of technology. Unsufficient tools, unsufficient knowledge. -
The story of GlobalSign • Really hard to troubleshoot •
So, hours wasted before the root cause is found • The fix must be immediate => cert management automation! -
Automated certificate management • CA with API • Let’s Encrypt?
Very good if you don’t need wildcard certificates. -
Automated certificate management • CA with API • Let’s Encrypt?
Very good if you don’t need wildcard certificates. • Tools like SSLMate • In-house plugins for ansible etc. -
What to set up during the deployment? • Strict Transport
Security • “Opportunistic encryption” simply doesn’t work • Most users won’t notice if HTTPS is absent • HTTPS only makes sense if it’s enforced -
What to set up during the deployment? • Strict Transport
Security • “Opportunistic encryption” simply doesn’t work • Most users won’t notice if HTTPS is absent • HTTPS only makes sense if it’s enforced • Public Key Pinning • Pin all end-entity public keys • Create a backup • Include future leafs • Rotate often => use automated tools to generate the header -
What to set up during the deployment? • Ciphers •
https://wiki.mozilla.org/Security/TLS_Configurations -
What to set up during the deployment? • Ciphers •
https://wiki.mozilla.org/Security/TLS_Configurations outdated • https://mozilla.github.io/server-side-tls/ssl-config-generator/ • Update frequently (automation?) -
What to set up during the deployment? • Ciphers •
https://wiki.mozilla.org/Security/TLS_Configurations outdated • https://mozilla.github.io/server-side-tls/ssl-config-generator/ • Update frequently (automation?) -
The story of Rijndael/AES • Ordered by U.S. federal government
• Approved by NSA, 1998-2001 • Adopted by U.S. DoD and Army -
The story of Rijndael/AES • Adopted by U.S. DoD and
Army • Military required three distinct security levels, with less sensitive data to be encrypted using the most weak method and vice versa -
The story of Rijndael/AES • Adopted by U.S. DoD and
Army • Military required three distinct security levels, with less sensitive data to be encrypted using the most weak method and vice versa • Crypto designers implemented three key sizes (128, 192, 256), with the most weak still unbreakable in foreseeable future (except quantum computers) -
The story of Rijndael/AES • Adopted by U.S. DoD and
Army • Military required three distinct security levels, with less sensitive data to be encrypted using the most weak method and vice versa • Crypto designers implemented three key sizes (128, 192, 256), with the most weak still unbreakable in foreseeable future (except quantum computers) • So, AES-128 is still good enough • Not that it matters much with modern AES-NI -
The story of Perfect Forward Secrecy • Present in ephemeral
Diffie-Hellman ciphers -
The story of Perfect Forward Secrecy • Present in ephemeral
Diffie-Hellman ciphers • Makes out-of-path analysis impossible • Makes historic data analysis impossible -
The story of Perfect Forward Secrecy • Present in ephemeral
Diffie-Hellman ciphers • Makes out-of-path analysis impossible • Makes historic data analysis impossible • Good catch for an out-of-path DPI and/or WAF 70% HTTPS requests come and go without analysis -
• Present in ephemeral Diffie-Hellman ciphers • Makes out-of-path analysis
impossible • Makes historic data analysis impossible • Good catch for an out-of-path DPI and/or WAF 70% HTTP requests go without analysis The story of Perfect Forward Secrecy 60% legitimate 90% malicious -
• Present in ephemeral Diffie-Hellman ciphers • Makes out-of-path analysis
impossible • Makes historic data analysis impossible • Good catch for an out-of-path DPI and/or WAF 70% HTTP requests go without analysis Oh boy, ciphers are tough. The story of Perfect Forward Secrecy 60% legitimate 90% malicious -
Protocols • SSLv2 is dead • SSLv3 is dead* •
TLSv1.0 is dead * – if you don’t have to serve content to IE6 or a TV set -
Protocols • SSLv2 is dead • SSLv3 is dead* •
TLSv1.0 is dead • TLS is alive and growing * – if you don’t have to serve content to IE6 or a TV set -
Protocols • SSLv2 is dead • SSLv3 is dead* •
TLSv1.0 is dead • TLS is alive and growing • Maybe too fast: TLSv1.2 allowed DDoSCoin * – if you don’t have to serve content to IE6 or a TV set -
Misc • OCSP stapling • Persistent connections (TLS handshake is
expensive) • Fight unencrypted content! -
Bonus track • Client certificates • May be combined with
2FA • May be integrated into certain applications as well • Unsupported by some mobile browsers OOTB :-(