--- 1/draft-ietf-tcpm-tcp-uto-02.txt 2006-07-20 01:12:24.000000000 +0200 +++ 2/draft-ietf-tcpm-tcp-uto-03.txt 2006-07-20 01:12:24.000000000 +0200 @@ -1,19 +1,19 @@ TCP Maintenance and Minor L. Eggert Extensions (tcpm) NEC Internet-Draft F. Gont -Expires: April 27, 2006 UTN/FRH - October 24, 2005 +Expires: January 20, 2007 UTN/FRH + July 19, 2006 TCP User Timeout Option - draft-ietf-tcpm-tcp-uto-02 + draft-ietf-tcpm-tcp-uto-03 Status of this Memo By submitting this Internet-Draft, each author represents that any applicable patent or other IPR claims of which he or she is aware have been or will be disclosed, and any of which he or she becomes aware will be disclosed, in accordance with Section 6 of BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that @@ -24,266 +24,253 @@ and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt. The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. - This Internet-Draft will expire on April 27, 2006. + This Internet-Draft will expire on January 20, 2007. Copyright Notice - Copyright (C) The Internet Society (2005). + Copyright (C) The Internet Society (2006). Abstract - The TCP user timeout controls how long transmitted data may remain - unacknowledged before a connection is forcefully closed. It is a - local, per-connection parameter. The advisory TCP User Timeout - Option allows conforming TCP implementations to exchange their local - user timeouts. This is an in-protocol mechanism to allow a host to - modify its local user timeout for a connection based on knowledge of - the peer's user timeout. Increasing the user timeouts allows - established TCP connections to survive extended periods of - disconnection. Decreasing user timeouts allows busy servers to - explicitly notify their clients that they will maintain the - connection state only across short periods of disconnection. + This document specifies a new TCP option - the TCP User Timeout + Option - that allows a TCP to advertise its current user timeout for + a connection. Thus, the remote TCP may modify its local user timeout + based on knowledge of the peer's user timeout. The TCP user timeout + controls how long transmitted data may remain unacknowledged before a + connection is forcefully closed. It is a local, per-connection + parameter. Increasing the user timeouts allows established TCP + connections to survive extended periods of disconnection. Decreasing + the user timeouts allows busy servers to explicitly notify their + clients that they will maintain the connection state only across + short periods of disconnection. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 2. Conventions . . . . . . . . . . . . . . . . . . . . . . . . . 4 - 3. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . 5 + 3. Operation . . . . . . . . . . . . . . . . . . . . . . . . . . 4 3.1. Changing the Local User Timeout . . . . . . . . . . . . . 6 3.2. Reliability Considerations . . . . . . . . . . . . . . . . 8 3.3. Option Format . . . . . . . . . . . . . . . . . . . . . . 8 3.4. Special Option Values . . . . . . . . . . . . . . . . . . 9 - 4. Additional Considerations . . . . . . . . . . . . . . . . . . 9 - 5. Interoperability Issues . . . . . . . . . . . . . . . . . . . 10 - 5.1. Middleboxes . . . . . . . . . . . . . . . . . . . . . . . 10 - 5.2. TCP Keep-Alives . . . . . . . . . . . . . . . . . . . . . 10 - 6. Security Considerations . . . . . . . . . . . . . . . . . . . 11 - 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 12 - 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 12 - 9. References . . . . . . . . . . . . . . . . . . . . . . . . . . 13 - 9.1. Normative References . . . . . . . . . . . . . . . . . . . 13 - 9.2. Informative References . . . . . . . . . . . . . . . . . . 13 + 4. Interoperability Issues . . . . . . . . . . . . . . . . . . . 9 + 4.1. Middleboxes . . . . . . . . . . . . . . . . . . . . . . . 9 + 4.2. TCP Keep-Alives . . . . . . . . . . . . . . . . . . . . . 10 + 5. Security Considerations . . . . . . . . . . . . . . . . . . . 10 + 6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 11 + 7. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 12 + 8. References . . . . . . . . . . . . . . . . . . . . . . . . . . 12 + 8.1. Normative References . . . . . . . . . . . . . . . . . . . 12 + 8.2. Informative References . . . . . . . . . . . . . . . . . . 12 Editorial Comments . . . . . . . . . . . . . . . . . . . . . . . . - Appendix A. Document Revision History . . . . . . . . . . . . . . 15 - Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 16 - Intellectual Property and Copyright Statements . . . . . . . . . . 17 + Appendix A. Alternative solutions . . . . . . . . . . . . . . . . 13 + Appendix B. Document Revision History . . . . . . . . . . . . . . 14 + Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 15 + Intellectual Property and Copyright Statements . . . . . . . . . . 16 1. Introduction - The Transmission Control Protocol (TCP) specification [RFC0793] - defines a local, per-connection "user timeout" parameter that - specifies the maximum amount of time that transmitted data may remain - unacknowledged before TCP will forcefully close the corresponding - connection. Applications can set and change this parameter with OPEN - and SEND calls. If a network disconnection lasts longer than the - user timeout, no acknowledgments will be received for any - transmission attempt, including keep-alives [TCP-ILLU], and the TCP - connection will close when the user timeout occurs. In the absence - of an application-specified user timeout, the TCP specification - [RFC0793] defines a default user timeout of 5 minutes. + The Transmission Control Protocol (TCP) specification + [RFC0793]defines a local, per-connection "user timeout" parameter + that specifies the maximum amount of time that transmitted data may + remain unacknowledged before TCP will forcefully close the + corresponding connection. Applications can set and change this + parameter with OPEN and SEND calls. If a network disconnection lasts + longer than the user timeout, no acknowledgments will be received for + any transmission attempt, including keep-alives [TCP-ILLU], and the + TCP connection will close when the user timeout occurs. In the + absence of an application-specified user timeout, the TCP + specification [RFC0793] defines a default user timeout of 5 minutes. The Host Requirements RFC [RFC1122] refines this definition by introducing two thresholds, R1 and R2 (R2 > R1), on the number of retransmissions of a single segment. It suggests that TCP should notify applications when R1 is reached for a segment, and close the connection once R2 is reached. [RFC1122] also defines the recommended values for R1 (three retransmissions) and R2 (100 seconds), noting that R2 for SYN segments should be at least 3 minutes. Instead of a single user timeout, some TCP implementations offer finer-grained policies. For example, Solaris supports different timeouts depending on whether a TCP connection is in the SYN-SENT, SYN-RECEIVED, or ESTABLISHED state [SOLARIS-MANUAL]. Although some TCP implementations allow applications to set their - local user timeout, e.g., through the SO_SNDTIMEO socket option, - there is no in-protocol mechanism to signal changes in the local user - timeout to remote peers. This causes local changes to be - ineffective, because the peer will still close the connection after - its user timeout expires, even when a host has raised its local user - timeout. The ability to modify the two user timeouts associated with - a connection can improve TCP operation in scenarios that are - currently not well supported. One example of such scenarios are - mobile hosts that change network attachment points based on current - location. Such hosts, maybe using MobileIP [RFC3344], HIP [I-D.ietf- - hip-arch] or transport-layer mobility mechanisms [I-D.eddy-tcp- - mobility], are only intermittently connected to the Internet. In - between connected periods, mobile hosts may experience periods of - disconnection during which no network service is available [SCHUETZ- - THESIS][SCHUETZ-CCR][DRIVE-THRU]. Other factors that can cause - transient periods of disconnection are high levels of congestion as - well as link or routing failures inside the network. + local user timeout, there is no in-protocol mechanism to signal + changes in the local user timeout to remote peers. This causes local + changes to be ineffective, because the peer will still close the + connection after its user timeout expires, even when the host has + raised its local user timeout. The ability to suggest the remote + peer a user timeout to be used for the connection can improve TCP's + operation in scenarios that are currently not well supported. One + example of such scenarios are mobile hosts that change network + attachment points based on current location. Such hosts, maybe using + MobileIP [RFC3344], HIP [RFC4423]or transport-layer mobility + mechanisms [I-D.eddy-tcp-mobility], are only intermittently connected + to the Internet. In between connected periods, mobile hosts may + experience periods of disconnection during which no network service + is available. Other factors that can cause transient periods of + disconnection are high levels of congestion as well as link or + routing failures inside the network. In scenarios similar to the ones described above, a host may not know exactly when or for how long it will be disconnected from the network, but it might expect such events due to past mobility patterns and thus benefit from using longer user timeouts. In other scenarios, the length and time of a network disconnection may even be - predictable. For example, an orbiting node on a satellite might - experience disconnections due to line-of-sight blocking by other - planetary bodies. The disconnection periods of such a node may be - easily computable from orbital mechanics. + predictable. For example, an orbiting node on a non-geostationary + satellite might experience disconnections due to line-of-sight + blocking by other planetary bodies. The disconnection periods of + such a node may be easily computable from orbital mechanics. This document specifies a new TCP option - the User Timeout Option - (UTO) - that allows conforming hosts to exchange their local, per- - connection user timeout information. This allows, for example, - mobile hosts to maintain TCP connections across disconnected periods - that are longer than their peer's default user timeout. A second use - of the TCP User Timeout Option is advertisement of shorter-than- - default user timeouts. This can allow busy servers to explicitly - notify their clients that they will maintain the state associated - with established connections only across short periods of - disconnection. + (UTO) - that allows a TCP to advertise its current local user timeout + parameter. Thus, based on the information advertised by the remote + TCP peer, a TCP may modify its own user timeout accordingly. This + allows, for example, mobile hosts to maintain TCP connections across + disconnected periods that are longer than their peer's default user + timeout. A second use of the TCP User Timeout Option is + advertisement of shorter-than-default user timeouts. This can allow + busy servers to explicitly notify their clients that they will + maintain the state associated with established connections only + across short periods of disconnection. - The same benefits can be obtained through an application-layer - mechanism, i.e., exchanging user timeout information via application - messages and having the application adjust the user timeouts through - the TCP API on both sides of a connection. This approach does not - require a new TCP option, but requires changing all application - implementations that desire to tolerate extended periods of - disconnection, and in most cases also requires a modification to the - corresponding application layer protocol. With the proposed TCP - option, application changes may not be necessary at all, or may be - restricted to sender- or receiver-side only, and there is no need to - modify the corresponding application protocol. + Use of the TCP User Timeout Option could be triggered either by an + API call or by a system-wide toggle. The API could be, for example, + a Socket option that would need to be explicitly set by the + corresponding application. This option would default to "off". A + system-wide toggle would allow a system administrator to enable the + use of the TCP User Timeout Option on a system-wide basis, and set + the option a desired value. This system-wide toggle would allow the + use of the option by application programs that have not been + explicitly coded to do so. If such a system-wide toggle were + provided, it would default to "off". - A different approach to tolerate longer periods of disconnection is - to simply increase the system-wide user timeout on both peers. This - approach has the benefit of not requiring a new TCP option or - application changes. However, it can also significantly increase the - amount of connection state a busy server must maintain, because a - longer global timeout value will apply to all its connections. The - proposed TCP User Timeout Option, on the other hand, allows hosts to - selectively manage the user timeouts of individual connections, - reducing the amount of state they must maintain across disconnected - periods. + In all cases, use of the TCP User Timeout Option would depend on an + active decision, either by the application programmer (by means of an + API call), or by a system administrator (by means of a system-wide + toggle). 2. Conventions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119]. 3. Operation Sending a TCP User Timeout Option informs the remote peer of the - current local user timeout and suggests that the remote peer SHOULD - start using the indicated user timeout value for the corresponding - connection. The user timeout value included in a TCP User Timeout - Option specifies the requested user timeout during the synchronized - states of a connection (ESTABLISHED, FIN-WAIT-1, FIN-WAIT-2, CLOSE- - WAIT, CLOSING, or LAST-ACK). Connections in other states MUST the - default timeout values defined in [RFC0793][RFC1122].[anchor3] + current local user timeout for the connection, and suggests the TCP + peer to adapt its user timeout accordingly. The user timeout value + included in a TCP User Timeout Option specifies the requested user + timeout during the synchronized states of a connection (ESTABLISHED, + FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, or LAST-ACK). + Connections in other states MUST the default timeout values defined + in [RFC0793] [RFC1122]. Note that an exchange of TCP User Timeout Options between peers is not a binding negotiation. Transmission of a TCP User Timeout Option is an advisory suggestion that the peer consider adapting its local - user timeout. Hosts remain free to forcefully close or abort - connections at any time for any reason, whether or not they use - custom user timeouts or have suggested the peer to use them. + user timeout. Hosts remain free to adopt a different user timeout, + or to forcefully close or abort connections at any time for any + reason, whether or not they use custom user timeouts or have + suggested the peer to use them. A host that supports the TCP User Timeout Option SHOULD include one in each packet that carries a SYN flag, but need not. [MEDINA] has shown that unknown options are correctly handled by the vast majority of modern TCP stacks. It is thus not necessary to require negotiation of the use of the TCP User Timeout Option during the - three-way handshake of a connection. + three-way handshake of a connection. However, including a TCP User + Timeout Option in each segment that has the SYN flag set will result + in TCP being able to adopt a user timeout with knowledge of that used + by the peer TCP, since the beginning of the data transfer phase. A host that supports the TCP User Timeout Option SHOULD include it in the next possible segment to its peer whenever it starts using a new user timeout for the connection. This allows the peer to adapt its local user timeout for the connection accordingly. When a host that supports the TCP User Timeout Option receives one, - it decides whether to change its local user timeout of the connection - based on the received value. Generally, hosts should honor requests - for changes to the user timeout (see Section 3.1), unless security - concerns, resource constraints or external policies indicate - otherwise (see Section 6). If so, hosts may ignore incoming TCP User - Timeout Options and use a different user timeout for the connection. - - When a host receives a TCP User Timeout Option, it first decides - whether to change its local user timeout for the connection - - Section 3.1 discusses the specifics of this choice - and then decides - whether to send a TCP User Timeout Option to its peer in response. - If a host has never sent a TCP User Timeout Option to its peer during - the lifetime of the connection, or if it has changed its local user - timeout, it SHOULD send TCP User Timeout Option with its current - local user timeout to its peer. [anchor4] + it will use the received value to compute the local user timeout for + the connection. Generally, hosts should honor requests for changes + to the user timeout (see Section 3.1), unless security concerns, + resource constraints or external policies indicate otherwise (see + Section 5). If so, hosts may use a different user timeout for the + connection. A TCP implementation that does not support the TCP User Timeout Option MUST silently ignore it [RFC1122], thus ensuring interoperability. - Hosts SHOULD impose upper and lower limits on the user timeouts they + Hosts MUST impose upper and lower limits on the user timeouts they use. Section 3.1 discusses user timeout limits, and describes a - recommended scheme to apply them. A TCP User Timeout Option with a + RECOMMENDED scheme to apply them. A TCP User Timeout Option with a value of zero (i.e., "now") is nonsensical and is used for a special purpose, see Section 3.4. Section 3.1 discusses potentially problematic effects of other user timeout durations. 3.1. Changing the Local User Timeout - When a host receives a TCP User Timeout Option, it MUST decide + When a host receives a TCP User Timeout Option, it must decide whether to change the local user timeout of the corresponding connection. Application-requested user timeout values always take precedence over timeout values received from the peer in a TCP User - Timeout Option. [anchor5] Consequently, unless the application on the + Timeout Option. [anchor3] Consequently, unless the application on the local host has requested a specific user timeout for the connection, - e.g., through a socket API call, hosts SHOULD adjust their local user - timeout in response to receiving a TCP User Timeout Option, as + e.g., through the OPEN or SEND calls, hosts SHOULD adjust their local + user timeout in response to receiving a TCP User Timeout Option, as described in the remainder of this section. If the local application has requested a specific local user timeout, TCP implementations MUST NOT change it in response to receiving a TCP User Timeout Option. In this case, they SHOULD, however, notify the application about the user timeout value received from the peer. The User Timeout Option specifies the user timeout in terms of time units, rather than in terms of number of retransmissions or round- trip times (RTTs), as in most cases the periods of disconnection have to do with operation and mobility patterns, rather than with the - current network conditions [anchor6][anchor7]. Thus, the TCP User - Timeout Option allows hosts to exchange user timeout values from 1 - second to over 9 hours at a granularity of seconds, and from 1 minute - to over 22 days at a granularity of minutes. (An option value of - zero is reserved for a special purpose, see Section 3.4.) + current network conditions. Thus, the TCP User Timeout Option allows + hosts to exchange user timeout values from 1 second to over 9 hours + at a granularity of seconds, and from 1 minute to over 22 days at a + granularity of minutes. (An option value of zero is reserved for a + special purpose, see Section 3.4.) Very short user timeout values can affect TCP transmissions over high-delay paths. If the user timeout occurs before an acknowledgment for an outstanding segment arrives, possibly due to packet loss, the connection closes. Many TCP implementations default to user timeout values of a few minutes [TCP-ILLU]. Although the TCP User Timeout Option allows suggestion of short timeouts, applications - advertising them SHOULD consider these effects. + advertising them should consider these effects. Long user timeout values allow hosts to tolerate extended periods of disconnection. However, they also require hosts to maintain the TCP state information associated with connections for long periods of - time. Section 6 discusses the security implications of long timeout + time. Section 5 discusses the security implications of long timeout values. - To protect against these effects, implementations SHOULD impose - limits on the user timeout values they accept and use. The remainder - of this section describes a RECOMMENDED scheme to limit user timeouts + To protect against these effects, implementations MUST impose limits + on the user timeout values they accept and use. The remainder of + this section describes a RECOMMENDED scheme to limit user timeouts based on upper and lower limits. Under the RECOMMENDED scheme, each TCP SHOULD compute the user timeout (USER_TIMEOUT) for a connection according to this formula: USER_TIMEOUT = min(U_LIMIT, max(LOCAL_UTO, REMOTE_UTO, L_LIMIT)) - Each field is to be interpreted as follows: USER_TIMEOUT Resulting user timeout value to be adopted by the local TCP for a connection. U_LIMIT Current upper limit imposed on the user timeout of a connection by the local host. @@ -291,86 +278,94 @@ Current lower limit imposed on the user timeout of a connection by the local host. LOCAL_UTO Current local user timeout of this specific connection. REMOTE_UTO Last "user timeout" value suggested by the remote peer by means of the TCP User Timeout Option. - This means that the maximum of the two announced values will be - adopted for the user timeout of the connection. The rationale is - that choosing the maximum of the two values will let the connection - survive longer periods of disconnection. If the TCP that announced - the lower of the two user timeout values did so in order to reduce - the amount of TCP state information that must be kept on the host, it - can, nevertheless, close or abort the connection whenever it wants. + This means that, provided they are within the upper and lower limits, + the maximum of the two announced values will be adopted for the user + timeout of the connection. The rationale is that choosing the + maximum of the two values will let the connection survive longer + periods of disconnection. If the TCP that announced the lower of the + two user timeout values did so in order to reduce the amount of TCP + state information that must be kept on the host, it can, + nevertheless, close or abort the connection whenever it wants. + + It must be noted that the two endpoints of the connection will not + necesarilly adopt the same user timeout. Enforcing a lower limit (L_LIMIT) prevents connections from closing due to transient network conditions, including temporary congestion, mobility hand-offs and routing instabilities. An upper limit (U_LIMIT) can reduce the effect of resource exhaustion - attacks. Section 6 discusses the details of these attacks. + attacks. Section 5discusses the details of these attacks. Note that these limits MAY be specified as system-wide constants or at other granularities, such as on per-host, per-user or even per- connection basis. Furthermore, these limits need not be static. For example, they MAY be a function of system resource utilization or attack status and could be dynamically adapted. The Host Requirements RFC [RFC1122] does not impose any limits on the length of the user timeout. However, a time interval of at least 100 seconds is RECOMMENDED. Consequently, the lower limit (L_LIMIT) SHOULD be set to at least 100 seconds when following the RECOMMENDED - scheme described in this section. + scheme described in this section. Adopting a user timeout smaller + than the current retransmission timeout (RTO) for the connection + would likely cause the connection to be aborted unnecesarily. + Therefore, the lower limit (L_LIMIT) MUST be larger than the current + retransmission timeout (RTO) for the connection. 3.2. Reliability Considerations The TCP User Timeout Option is an advisory TCP option that does not change processing of subsequent segments. Unlike other TCP options, it need not be exchanged reliably. Consequently, the specification in this section does not define a reliability handshake for TCP User Timeout Option exchanges. When a segment that carries a TCP User Timeout Option is lost, the option may never reach the intended peer. Implementations MAY implement local mechanisms to improve delivery reliability, such as retransmitting the TCP User Timeout Option when they retransmit the segment that originally carried it, or "attaching" the option to a byte in the stream and retransmitting the option whenever that byte or its ACK are retransmitted. It is important to note that although these mechanisms can improve transmission reliability for the TCP User Timeout Option, they do not guarantee delivery (a three-way handshake would be required for - this). Consequently, implementations MUST NOT assume that a TCP User - Timeout Option is reliably transmitted. + this). Consequently, implementations should not assume that a TCP + User Timeout Option is reliably transmitted. 3.3. Option Format 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Kind = X | Length = 4 |G| User Timeout | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ (One tick mark represents one bit.) Figure 1: Format of the TCP User Timeout Option Figure 1 shows the format of the TCP User Timeout Option. It contains these fields: Kind (8 bits) A TCP option number [RFC0793] to be assigned by IANA upon - publication of this document (see Section 7). + publication of this document (see Section 6). Length (8 bits) Length of the TCP option in octets [RFC0793]; its value MUST be 4. Granularity (1 bit) Granularity bit, indicating the granularity of the "User Timeout" field. When set (G = 1), the time interval in the "User Timeout" field MUST be interpreted as minutes. Otherwise (G = 0), the time interval in the "User Timeout" field MUST be interpreted as seconds. @@ -397,105 +392,81 @@ calculating a new local USER_TIMEOUT described in Section 3.1, with a numeric value of zero seconds for REMOTE_UTO. The sender SHOULD perform the same calculation as described in Section 3.1, with a numeric value of zero seconds for LOCAL_UTO. A zero-minute TCP User Timeout Option, i.e., with a "User Timeout" field of zero and a "Granularity" bit of one, is reserved for future use. TCP implementations MUST NOT send it and MUST ignore it upon reception. -4. Additional Considerations - - Section 1 described that although [RFC0793] defines the API mechanism - to change the user timeout as an optional parameter for TCP's OPEN - and SEND calls, many implementations provide a different API. - Several popular TCP implementations offer the SO_SNDTIMEO socket - option, either in addition or instead of the RFC-defined OPEN and - SEND user timeout parameter. - - Many implementations that offer the SO_SNDTIMEO socket option also - implement a corresponding SO_RCVTIMEO socket option. Whereas the - user timout (SO_SNDTIMEO), specifies how long data may remain - unacknowledged by the peer, i.e., how long a SEND call may take, the - SO_RCVTIMEO specifies how long a RECV call may take. - - Even when two TCPs implement the TCP User Timeout Option and decide - to lengthen their local UTOs for a connection, RECV operations during - a disconnection can trigger the SO_RCVTIMEO timeout. Note that - [RFC0793] does not specify this receive timeout or how TCP reacts - when it occurs. If implementations close a connection when its - SO_RCVTIMEO times out, they SHOULD modify this parameter similarly to - how they modify SO_SNDTIMEO upon reception of a TCP User Timeout - option. [anchor10] - -5. Interoperability Issues +4. Interoperability Issues This section discusses interoperability issues related to introducing the TCP User Timeout Option. -5.1. Middleboxes +4.1. Middleboxes The large number of middleboxes (firewalls, proxies, protocol scrubbers, etc.) currently present in the Internet pose some difficulty for deploying new TCP options. Some firewalls may block segments that carry unknown options, preventing connection - establishment when the SYN or SYN-ACK contains a TCP User Timeout - Option. Some recent results, however, indicate that for new TCP - options, this may not be a significant threat, with only 0.2% of web - requests failing when carrying an unknown option [MEDINA]. + establishment when the SYN or SYN-ACK segment contain a TCP User + Timeout Option. Some recent results, however, indicate that for new + TCP options, this may not be a significant threat, with only 0.2% of + web requests failing when carrying an unknown option [MEDINA]. Stateful firewalls usually reset connections after a period of inactivity. If such a firewall exists along the path between two peers, it may close or abort connections regardless of the use of the TCP User Timeout Option. In the future, such firewalls may learn to parse the TCP User Timeout Option and modify their behavior or the option accordingly. -5.2. TCP Keep-Alives +4.2. TCP Keep-Alives Some TCP implementations, such as the one in BSD systems, use a different abort policy for TCP keep-alives than for user data. Thus, the TCP keep-alive mechanism might abort a connection that would otherwise have survived the transient period of disconnection. Therefore, if a TCP peer enables TCP keep-alives for a connection that is using the TCP User Timeout Option, then the keep-alive timer MUST be set to a value larger than that of the adopted USER TIMEOUT. -6. Security Considerations +5. Security Considerations Lengthening user timeouts has obvious security implications. Flooding attacks cause denial of service by forcing servers to commit resources for maintaining the state of throw-away connections. However, TCP implementations do not become more vulnerable to simple SYN flooding by implementing the TCP User Timeout Option, because user timeouts exchanged during the handshake only affect the synchronized states (ESTABLISHED, FIN-WAIT-1, FIN-WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK), which simple SYN floods never reach. However, when an attacker completes the three-way handshakes of its throw-away connections it can amplify the effects of resource exhaustion attacks, because the attacked server must maintain the connection state associated with the throw-away connections for longer durations. Because connection state is kept longer, lower- frequency attack traffic, which may be more difficult to detect, can already cause resource exhaustion. Several approaches can help mitigate this issue. First, implementations can require prior peer authentication, e.g., using - IPsec [I-D.ietf-ipsec-rfc2401bis], before accepting long user - timeouts for the peer's connections. Similarly, a host can start to - accept long user timeouts for an established connection only after - in-band authentication has occurred, for example, after a TLS - handshake across the connection has succeeded [RFC2246]. Although - these are arguably the most complete solutions, they depend on - external mechanisms to establish a trust relationship. + IPsec [RFC4301], before accepting long user timeouts for the peer's + connections. Similarly, a host can start to accept long user + timeouts for an established connection only after in-band + authentication has occurred, for example, after a TLS handshake + across the connection has succeeded [RFC2246]. Although these are + arguably the most complete solutions, they depend on external + mechanisms to establish a trust relationship. A second alternative that does not depend on external mechanisms would introduce a per-peer limit on the number of connections that may use increased user timeouts. Several variants of this approach are possible, such as fixed limits or shortening accepted user timeouts with a rising number of connections. Although this alternative does not eliminate resource exhaustion attacks from a single peer, it can limit their effects. Reducing the number of high-UTO connections a server supports in the face of an attack turns that attack into a denial-of-service attack against the service of @@ -516,163 +487,142 @@ effective. In general, using the peers' level of trust as a parameter during the load-shedding decision process may be useful. Note that if TCP needs to close or abort connections with a long TCP User Timeout Option to shed load, these connections are still no worse off than without the option. Finally, upper and lower limits on user timeouts, discussed in Section 3.1, can be an effective tool to limit the impact of these sorts of attacks. -7. IANA Considerations +6. IANA Considerations This section is to be interpreted according to [RFC2434]. This document does not define any new namespaces. It uses an 8-bit TCP option number maintained by IANA at http://www.iana.org/assignments/tcp-parameters. -8. Acknowledgments +7. Acknowledgments The following people have improved this document through thoughtful - suggestions: Mark Allmann, David Borman, Bob Braden, Marcus Brunner, - Wesley Eddy, Abolade Gbadegesin, Ted Faber, Guillermo Gont, Tom - Henderson, Joseph Ishac, Jeremy Harris, Phil Karn, Michael Kerrisk, - Dan Krejsa, Kostas Pentikousis, Juergen Quittek, Joe Touch, Stefan - Schmid, Simon Schuetz, Tim Shepard and Martin Stiemerling. + suggestions: Mark Allman, David Borman, Bob Braden, Marcus Brunner, + Wesley Eddy, Gorry Fairhurst, Abolade Gbadegesin, Ted Faber, + Guillermo Gont, Tom Henderson, Joseph Ishac, Jeremy Harris, Phil + Karn, Michael Kerrisk, Dan Krejsa, Kostas Pentikousis, Juergen + Quittek, Joe Touch, Stefan Schmid, Simon Schuetz, Tim Shepard and + Martin Stiemerling. Lars Eggert is partly funded by Ambient Networks, a research project supported by the European Commission under its Sixth Framework Program. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the Ambient Networks project or the European Commission. -9. References +8. References -9.1. Normative References +8.1. Normative References [RFC0793] Postel, J., "Transmission Control Protocol", STD 7, RFC 793, September 1981. [RFC1122] Braden, R., "Requirements for Internet Hosts - Communication Layers", STD 3, RFC 1122, October 1989. [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC2434] Narten, T. and H. Alvestrand, "Guidelines for Writing an IANA Considerations Section in RFCs", BCP 26, RFC 2434, October 1998. -9.2. Informative References - - [DRIVE-THRU] - Ott, J. and D. Kutscher, "Drive-Thru Internet: IEEE - 802.11b for Automobile Users", Proc. Infocom , March 2004. +8.2. Informative References [I-D.eddy-tcp-mobility] Eddy, W., "Mobility Support For TCP", draft-eddy-tcp-mobility-00 (work in progress), April 2004. - [I-D.ietf-hip-arch] - Moskowitz, R. and P. Nikander, "Host Identity Protocol - Architecture", draft-ietf-hip-arch-03 (work in progress), - August 2005. - - [I-D.ietf-ipsec-rfc2401bis] - Kent, S. and K. Seo, "Security Architecture for the - Internet Protocol", draft-ietf-ipsec-rfc2401bis-06 (work - in progress), April 2005. - [MEDINA] Medina, A., Allman, M., and S. Floyd, "Measuring Interactions Between Transport Protocols and Middleboxes", Proc. 4th ACM SIGCOMM/USENIX Conference on Internet Measurement , October 2004. [RFC2246] Dierks, T. and C. Allen, "The TLS Protocol Version 1.0", RFC 2246, January 1999. [RFC3344] Perkins, C., "IP Mobility Support for IPv4", RFC 3344, August 2002. - [SCHUETZ-CCR] - Schuetz, S., Eggert, L., Schmid, S., and M. Brunner, - "Protocol Enhancements for Intermittently Connected - Hosts", To appear: ACM Computer Communication Review, Vol. - 35, No. 3, July 2005. + [RFC4301] Kent, S. and K. Seo, "Security Architecture for the + Internet Protocol", RFC 4301, December 2005. - [SCHUETZ-THESIS] - Schuetz, S., "Network Support for Intermittently Connected - Mobile Nodes", Diploma Thesis, University of Mannheim, - Germany, June 2004. + [RFC4423] Moskowitz, R. and P. Nikander, "Host Identity Protocol + (HIP) Architecture", RFC 4423, May 2006. [SOLARIS-MANUAL] Sun Microsystems, "Solaris Tunable Parameters Reference Manual", Part No. 806-7009-10, 2002. [TCP-ILLU] Stevens, W., "TCP/IP Illustrated, Volume 1: The Protocols", Addison-Wesley , 1994. Editorial Comments - [anchor3] A future version of this document may extend per- - connection user timeouts to the SYN-SENT and SYN-RECEIVED - states in a way that conforms to the required minimum - timeouts. - - [anchor4] Should it really always send UTO when it changes the - local timeout? I can imagine some ping-pong effect when - two hosts user different UTO adoption strategies. But - maybe that's OK? Additionally, when -01 was presented in - Paris, Joe Touch has suggested that an "UTO-ACK" should - be sent when a UTO is received. I have not seen consensus - for this on the mailing list, hence -02 does not include - this suggestion. - - [anchor5] Without this, UTO would modify TCP semantics, because + [anchor3] Without this, UTO would modify TCP semantics, because application-requested UTOs could be overridden by peer requests. - [anchor6] When -01 was presented in Paris, Bob Braden suggested to - specify the UTO in terms of multiples of the RTT. Others - disagreed, hence -02 does not include this suggestion. - One reason this may be problematic is that the RTT may - change for one direction of the connection, sort of - defeating the process of exchanging UTOs. +Appendix A. Alternative solutions - [anchor7] Let's suppose a host is intermittently connected to a - network, and the disconnected periods last for, say, - about 15 minutes each. Let's suppose the attachment - points range from low-bandwidth/high-delay/congested - networks to high-bandwidth/low-delay networks. If the UTO - specified the user timeout in terms of number of - retransmissions or round-trip times, an UTO that is - appropriate for the high-bandwith/low-delay networks - would be too small for the low-bandwidth/high-delay/ - congested networks. Also, even if the host kept connected - to the same network, if the network conditions changed, - UTO opions would need to be re-sent (as n*RTO and n*RTT - would change), unnecesarily. + The same benefits could be obtained through an application-layer + mechanism, i.e., exchanging user timeout information via application + messages and having the application adjust the user timeouts through + the TCP API on both sides of a connection. This approach would not + require a new TCP option, but would require changing all application + implementations that desire to tolerate extended periods of + disconnection, and in most cases would also require a modification to + the corresponding application layer protocol. With the proposed TCP + option, application changes may not be necessary at all, or may be + restricted to sender- or receiver-side only, and there is no need to + modify the corresponding application protocol. - [anchor10] Can we even say this much about an API that's not in the - TCP spec? Or should the SO_RCVTIMEO discussion be - removed? + A different approach to tolerate longer periods of disconnection + would be to simply increase the system-wide user timeout on both + peers. This approach has the benefit of not requiring a new TCP + option or application changes. However, it can also significantly + increase the amount of connection state a busy server must maintain, + because a longer global timeout value would apply to all its + connections. -Appendix A. Document Revision History + The proposed TCP User Timeout Option, on the other hand, allows hosts + to selectively manage the user timeouts of individual connections, + reducing the amount of state they must maintain across disconnected + periods. + +Appendix B. Document Revision History To be removed upon publication +----------+--------------------------------------------------------+ | Revision | Comments | +----------+--------------------------------------------------------+ + | 03 | Corrected use of RFC2119 terminology. Clarified how | + | | use of the TCP UTO is triggered. Clarified reason for | + | | sending a UTO in the SYN and SYN/ACK segments. | + | | Removed discussion of the SO_SNDTIMEO and SO_RCVTIMEO | + | | options. Removed text that suggested that a UTO | + | | should be sent upon receipt of an UTO from the remote | + | | peer. Required minimum value for the lower limit of | + | | the user timeout. Moved alternative solutions to | + | | appendix. Miscellaneous editorial changes. | | 02 | Corrected terminology by replacing terms like | | | "negotiate", "coordinate", etc. that were left from | | | pre-WG-document times when the UTO was a more | | | formalized exchange instead of the advisory one it is | | | now. Application-requested UTOs take precedence over | | | ones received from the peer (pointed out by Ted | | | Faber). Added a brief mention of SO_SNDTIMEO and a | | | slightly longer discussion of SO_RCVTIMEO. | | 01 | Clarified and corrected the description of the | | | existing user timeout in RFC793 and RFC1122. Removed | @@ -741,18 +691,18 @@ This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Copyright Statement - Copyright (C) The Internet Society (2005). This document is subject + Copyright (C) The Internet Society (2006). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. Acknowledgment Funding for the RFC Editor function is currently provided by the Internet Society.