Strict Transport Security

Copyright © 2015-2017 Attila Molnar <>

Copyright © 2016-2017 James Wheare <>

Unlimited redistribution and modification of this document is allowed provided that the above copyright notice and this permission notice remains intact.

Description ๐Ÿ”—

Strict Transport Security (STS) is a mechanism which allows servers to advertise a policy that clients should only connect to them over a secure connection.

The policy is communicated to clients via the STS capability and should be processed by the client at capability negotiation time.

The name of the STS capability is sts.

The value of the capability specifies the duration during which the client MUST only connect to the server securely (using TLS, aka. SSL), and the port number to use for secure connections.

Clients MUST use sufficient certificate verification to ensure the connection is secure without any errors. This might involve validation against a system root certificate bundle or a user-specified trust root.

In server software, this capability MUST be made available as an optional configuration setting. Server administrators might have valid reasons not to enable it.

Details ๐Ÿ”—

When enabled, the capability has a REQUIRED value: a comma (,) (0x2C) separated list of tokens. Each token consists of a key which might have a value attached. If there is a value attached, the value is separated from the key by an equals sign (=) (0x3D). That is, <key>[=<value>][,<key2>[=<value2>][,<keyN>[=<valueN>]]]. Keys specified in this document MUST only occur at most once.

Clients MUST ignore every token with a key that they donโ€™t understand.

An STS policy has several parts:

If any required part is missing from the STS policy, clients MUST continue as if no STS policy was advertised.

Servers MAY advertise all STS policy parts together on both secure and insecure connections. Clients MUST only respect each policy part on the appropriate connection type.

See the capability negotiation specification for more information about capabilities with values.

Mechanism ๐Ÿ”—

When a client sees an STS upgrade policy over an insecure connection, it MUST first establish a secure connection (see the port key) and confirm that the STS persistence policy is present.

Once a client has connected securely, and it has verified that an STS persistence policy is in place, it then MUST only use a secure transport to connect to the server at the requested hostname in future, until the policy expires (see the duration key). Once an STS persistence policy has been verified, clients MUST refuse to connect if a secure connection cannot be established to the server for any reason during the lifetime of the policy.

If the secure connection succeeds but an STS persistence policy is not present, the client SHOULD continue using the secure connection for that session. This allows servers to upgrade client connections without committing to a more permanent STS policy.

Clients MUST NOT request this capability with CAP REQ. Servers MAY reply with a CAP NAK message if a client requests this capability.

Servers MAY communicate changes to their STS persistence policy using the CAP NEW command provided by cap-notify or CAP Version 302. Clients MUST store or update an STS policy for the hostname of a securely connected server if they receive a new STS capability notification.

Servers MUST NOT send CAP DEL to disable this capability, and clients MUST ignore any attempts to do so. The mechanism for disabling an STS persistence policy is described in the duration key section.

The port key ๐Ÿ”—

This key indicates the port number for making a secure connection. This keyโ€™s value MUST be a single port number.

If the client is not already connected securely to the server at the requested hostname, it MUST close the insecure connection and reconnect securely on the stated port.

To enforce an STS upgrade policy, servers MUST send this key to insecurely connected clients. Servers MAY send this key to securely connected clients, but it will be ignored.

The duration key ๐Ÿ”—

This key is used on secure connections to indicate how long clients MUST continue to use secure connections when connecting to the server at the requested hostname. The value of this key MUST be given as a single integer which represents the number of seconds until the persistence policy expires.

To enforce an STS persistence policy, servers MUST send this key to securely connected clients. Servers MAY send this key to all clients, but insecurely connected clients MUST ignore it.

Clients MUST reset the STS policy expiry time for a requested hostname every time a valid persistence policy with a duration key is received on a secure connection.

A duration=0 value indicates a disabled STS persistence policy with an immediate expiry. This value MAY be used by servers to remove a previously set policy.

When an STS persistence policy expires, clients MAY continue to connect securely to the server at the requested hostname, but they are no longer required to pre-emptively upgrade insecure connections. Clients MUST still respect any STS upgrade policies encountered when a persistence policy is expired or disabled.

The preload key ๐Ÿ”—

This OPTIONAL key, if present on a secure connection, indicates that the server agrees to be included in STS preload lists. If it has a value, the value MUST be ignored.

Servers MAY send this key to all clients, but insecurely connected clients MUST ignore it.

Preload list providers MUST only consider hosts for inclusion after validating their connection security and ensuring a valid STS policy with a preload key is in place. This allows IRC network administrators to opt-in for inclusion in preload lists.

Servers SHOULD be prepared to offer secure connections for the long term when enabling a preload policy. Timely removal of hostnames from preload lists might not be possible.

Preload list providers SHOULD consider STS persistence policy durations and MAY set minimum duration requirements prior to inclusion. Clients using preload lists SHOULD consider how their release cycle compares to any duration requirements imposed by list providers.

Server Name Indication ๐Ÿ”—

Before advertising an STS persistence policy over a secure connection, servers SHOULD verify whether the hostname provided by clients, for example, via TLS Server Name Indication (SNI), has been whitelisted by administrators in the server configuration.

If no hostname has been provided for the connection, an STS persistence policy SHOULD NOT be advertised.

This allows server administrators to retain control over which hostnames are STS-enabled in case the server is accessible on multiple hostnames. It is possible that a server uses a wildcard certificate or a certificate with Subject Alternative Names but its administrators only wish to advertise STS on a subset of its hostnames.

Take for example a server presenting a wildcard certificate for * The hostnames,, and all point to the same IP address. The server administrators may only wish to have STS enabled for, but no other hostname.

IRCd software SHOULD allow for each part of the STS policy to be configured per hostname. This allows server administrators to, for example, enable STS persistence on all hostnames, but only enable a preload policy for a subset of them.

SNI hostname verification is not available on insecure connections, so it might not be possible to configure a variable upgrade policy for multiple hostnames that share an IP address.

Rescheduling expiry on disconnect ๐Ÿ”—

IRC connections can be long-lived. Connections that last for more than a month are not uncommon. When a client activates an STS persistence policy for a hostname on a long-lived connection, the expiry time might be reached by the time the connection closes. However, the server might still have an STS policy in place.

To avoid an early STS policy expiry, clients MUST reschedule the expiry time when closing connections. The new expiry time is calculated by adding the policy duration as last advertised by the server to the time the connection is closed.

General Security considerations ๐Ÿ”—

This section is non-normative.

STS policy stripping ๐Ÿ”—

Itโ€™s possible for an attacker to remove the STS port value from an initial connection established via an insecure connection, before the policy has been cached by the client. This represents a bootstrap MITM (man-in-the-middle) vulnerability.

Clients might choose to mitigate this risk by implementing features such as user-declared and preloaded STS policies.

Denial of Service ๐Ÿ”—

STS could result in a Denial of Service (DoS) on IRC servers in a number of ways. A non-exhaustive list of examples might include:

These issues are not vulnerabilities with STS itself, but rather are compounding issues for configuration errors, or issues involving vulnerable systems exploited by other means.

Clients and servers can mitigate many of these issues by adopting mechanisms such as preload lists described in the following implementation consideration sections

Client implementation considerations ๐Ÿ”—

This section is non-normative.

For increased user protection and more advanced management of cached STS policies, clients might consider implementing features such as the following:

Rescheduling while connected ๐Ÿ”—

In addition to rescheduling expiry on disconnect, clients might choose to periodically reschedule policy expiry while connected as well. This could provide extra protection in the case of a sudden power failure, for example.

An appropriate period for rescheduling expiry while connected might depend on the policyโ€™s duration. For instance, a longer policy duration might warrant less frequent rescheduling.

No immediate user recourse ๐Ÿ”—

If a client fails to establish a secure connection to a hostname with an active STS policy, there should be no immediate user recourse to bypass the failure. Clients should treat a failure to establish a secure connection to an STS-enabled host as a critical error, and shouldnโ€™t offer a prominent way to retry it insecurely. For example, the user should not be presented with a dialog that allows them to proceed with the connection with a click. Users should instead be required to retry the connection manually at a later time.

This advice is intended to avoid teaching users that strict security errors can be ignored, and to offer an extra layer of protection from man-in-the-middle attacks.

User-declared STS policy ๐Ÿ”—

Clients might consider allowing users to explicitly define an STS policy for a given host, before any interaction with the host. This could help prevent a bootstrap MITM vulnerability as discussed in the STS policy stripping section.

STS policy deletion or rejection ๐Ÿ”—

Clients might consider allowing users or administrators to reject or remove cached STS policies on a per-host basis, in case a serverโ€™s policy is accidentally or maliciously injected on a secure connection.

Such a feature should be made available very carefully from both user interface and security standpoints. Deleting or rejecting a cache entry for a known STS host should be a very deliberate and well-considered act โ€“ it shouldnโ€™t be something that users get used to doing as a matter of course: e.g., just โ€œclicking throughโ€ in order to get work done. In other words, these features should not violate the no immediate user recourse section.

Server implementation considerations ๐Ÿ”—

This section is non-normative.

Policy expiry time ๐Ÿ”—

Network administrators should consider whether their policy duration represents a constant value into the future, or a fixed expiry time.

Constant values into the future can be achieved by a configured number of seconds being sent in the duration key on each connection attempt.

Fixed expiry times will involve a dynamic duration value being calculated on each connection attempt.

Server implementers should be aware that fixed expiry times might not be precisely guaranteed in the case where clients reschedule policy expiry on disconnect, or periodically while connected.

Which approach to take will depend on a number of considerations. For example, a server might wish their STS Policy to expire at the same time as their hostname certificate. Alternatively, a server might wish their STS policy to last for as long as possible.

Server administrators should be aware of the ability to set a duration=0 value at any time to revoke an STS policy if secure connections are no longer enabled for their servers.

Server implementations should consider using a default value of duration=0 in their example configurations. This will require server administrators to deliberately choose an expiry according to their specific needs rather than (perhaps unknowingly) rely on an arbitrary generic value.

Offering multiple IRC servers at alternate ports on the same hostname ๐Ÿ”—

An STS policy is imposed for all connections to a hostname. This can create issues when attempting to run multiple IRC servers that share a hostname but use different ports.

For example, a single host might run a production IRC server which advertises an STS policy and another, insecure IRC server on a different port for testing purposes. STS capable clients will lose the ability to connect to the testing server once an STS policy is established for the production server.

To avoid these issues, servers can be offered on separate hostnames, perhaps using subdomains.

Relationship with other specifications ๐Ÿ”—

This section is non-normative.

Relationship with STARTTLS ๐Ÿ”—

STARTTLS is a mechanism for upgrading a connection which has started out as an insecure connection to a secure connection without reconnecting to a different port. This means a server can offer both insecure and secure connections on the same port for compatible clients at the cost of more complex implementations in both clients and servers.

In practice, switching protocols in the middle of the stream has proven to be complicated enough that only a small number of clients bothered implementing STARTTLS.

STS expects that servers instead offer a port that directly services secure connections and it is incompatible with servers that offer secure connections only via STARTTLS on an insecure port.

Relationship with the tls capability ๐Ÿ”—

The tls capability is a hint for clients that secure connection support is available via STARTTLS.

STS is an improved solution and should be considered a replacement for multiple reasons.

Examples ๐Ÿ”—

Redirecting to a secure port with an upgrade policy ๐Ÿ”—

A server tells a client connecting insecurely to connect securely on port 6697.

  Insecure Client: CAP LS 302
           Server: CAP * LS :sts=port=6697

After the exchange, the client disconnects and reconnects securely to the same hostname on port 6697.

Setting an STS persistence policy with a duration ๐Ÿ”—

A server tells a secure client to only use secure connections for roughly 6 months.

Secure Client: CAP LS 302
       Server: CAP * LS :sts=duration=15552000

Until the policy expires:

Ignoring an invalid request ๐Ÿ”—

A server tells an insecure client to use secure connections for roughly 6 months. There is no port advertised.

  Insecure Client: CAP LS 302
           Server: CAP * LS :sts=duration=15552000

The client ignores this because it has received an STS persistence policy over an insecure connection and the STS cap doesnโ€™t contain an upgrade policy.

Handling tokens with unknown keys ๐Ÿ”—

A server tells a secure client to use secure connections for roughly a year, but the value of the STS capability also contains tokens the client doesnโ€™t recognise.

Secure Client: CAP LS 302
       Server: CAP * LS :sts=unknown,duration=31536000,foo=bar

The client ignores the keys it does not understand and until the policy expires:

Handling 0 duration ๐Ÿ”—

A server tells a client that is already connected securely to remove the STS policy immediately.

Secure Client: CAP LS 302
       Server: CAP * LS :sts=duration=0

If the client has an STS policy stored for the host it clears the policy. Future attempts to connect insecurely will be allowed.

Rescheduling expiry on disconnect ๐Ÿ”—

A client securely connects to a server, which advertises an STS policy.

Secure Client: CAP LS 302
       Server: CAP * LS :multi-prefix sts=duration=2592000

The client saves the policy and notes that it will expire in 2592000 seconds (roughly one month). It completes registration, then proceeds as usual.

After 48 hours, the client disconnects.

Secure Client: QUIT :Bye

The policy is still valid, so the client reschedules expiry for 2592000 seconds from the time of disconnection.

Updating an STS policy with CAP NEW ๐Ÿ”—

A server updates an sts policy by sending a CAP NEW notification.

Server: CAP * NEW :sts=duration=31536000

If the client has an STS policy stored for the host it updates the policy to expire after 31536000 seconds. Otherwise a new policy is saved for the server.

Removing an STS policy with CAP NEW ๐Ÿ”—

A server removes an sts policy by sending a CAP NEW notification.

Server: CAP * NEW :sts=duration=0

If the client has an STS policy stored for the host it clears the policy. Future attempts to connect insecurely will be allowed.

Receiving CAP DEL ๐Ÿ”—

A server makes an invalid attempt to remove an sts policy by disabling the CAP.

Server: CAP * DEL :sts

The client will ignore this attempt and only rely on receiving a duration of 0 to disable STS policies.

Enabling an STS preload policy ๐Ÿ”—

A client securely connects to a server, which advertises an STS policy and opts in to preload lists.

Secure Client: CAP LS 302
       Server: CAP * LS :sts=duration=2592000,preload

Software supporting sts: Ergo, IRCCloud Teams, InspIRCd, UnrealIRCd, AdiIRC, Ambassador, Glirc, IceChat, mIRC, IRCCloud, The Lounge, Colloquy, Palaver, IRCCloud (as Server), pounce (as Server), BitBot, Limnoria, girc, ircrobots, Kitteh IRC Client Library