Network Working Group D. Cridland Request for Comments: 5267 C. King Category: Standards Track Isode Limited July 2008
Contexts for IMAP4
Status of This Memo
This document specifies an Internet standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "Internet Official Protocol Standards" (STD 1) for the standardization state and status of this protocol. Distribution of this memo is unlimited.
Abstract
The IMAP4rev1 protocol has powerful search facilities as part of the core protocol, but lacks the ability to create live, updated results that can be easily handled. This memo provides such an extension, and shows how it can be used to provide a facility similar to virtual mailboxes.
Although the basic SEARCH command defined in [IMAP], and as enhanced by [ESEARCH], is relatively compact in its representation, this reduction saves only a certain amount of data, and huge mailboxes might overwhelm the storage available for results on even relatively high-end desktop machines.
The SORT command defined in [SORT] provides useful features, but is hard to use effectively on changing mailboxes over low-bandwidth connections.
This memo borrows concepts from [ACAP], such as providing a windowed view onto search or sort results, and making updates that are bandwidth and round-trip efficient. These are provided by two extensions: "ESORT" and "CONTEXT".
In examples, "C:" and "S:" indicate lines sent by the client messaging user agent and IMAP4rev1 ([IMAP]) server, respectively. "//" indicates inline comments not part of the protocol exchange. Line breaks are liberally inserted for clarity. Examples are intended to be read in order, such that the state remains from one example to the next.
Although the examples show a server that supports [ESEARCH], this is not a strict requirement of this specification.
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 [KEYWORDS].
Other capitalized words are typically names of IMAP extensions or commands -- these are uppercased for clarity only, and are case- insensitive.
Servers implementing the extended SORT provide a suite of extensions to the SORT and UID SORT commands defined in [SORT]. This allows for return options, as used with SEARCH and specified in [IMAP-ABNF], to be used with SORT in a similar manner.
The SORT and UID SORT commands are extended by the addition of an optional list of return options that follow a RETURN atom immediately after the command. If this is missing, the server will return results as specified in [SORT].
Cridland & King Standards Track [Page 3]
RFC 5267 IMAP CONTEXT July 2008
The extended SORT command always returns results in the requested sort order, but is otherwise identical in its behaviour to the extended SEARCH command defined in [IMAP-ABNF], as extended by [ESEARCH]. In particular, the extended SORT command returns results in an ESEARCH response.
Servers advertising the capability "ESORT" support the return options specified in [ESEARCH] in the SORT command. These return options are adapted as follows:
MIN Return the message number/UID of the lowest sorted message satisfying the search criteria.
MAX Return the message number/UID of the highest sorted message satisfying the search criteria.
ALL Return all message numbers/UIDs which match the search criteria, in the requested sort order, using a sequence-set. Note the use of ranges described below in Section 3.2.
Any ranges given by the server, including those given as part of the sequence-set, in an ESEARCH response resulting from an extended SORT or UID SORT command, MUST be ordered in increasing numerical order after expansion, as per usual [IMAP] rules.
In particular this means that 10:12 is equivalent to 12:10, and 10,11,12. To avoid confusion, servers SHOULD present ranges only when the first seq-number is lower than the second; that is, either of the forms 10:12 or 10,11,12 is acceptable, but 12:10 SHOULD be avoided.
If the list of return options is present but empty, then the server provides the ALL return data item in an ESEARCH response. This is functionally equivalent to an unextended UID SORT command, but can use a smaller representation:
The Contexts extension is present in any IMAP4rev1 server that includes the string "CONTEXT=SEARCH", and/or "CONTEXT=SORT", within its advertised capabilities.
In the case of CONTEXT=SEARCH, the server supports the extended SEARCH command syntax described in [IMAP-ABNF], and accepts three additional return options.
Servers advertising CONTEXT=SORT also advertise the SORT capability, as described in [SORT], support the extended SORT command syntax described in Section 3, and accept three additional return options for this extended SORT.
These additional return options allow for notifications of changes to the results of SEARCH or SORT commands, and also allow for access to partial results.
A server advertising the CONTEXT=SEARCH extension will order all SEARCH results, whether from a UID SEARCH or SEARCH command, in mailbox order -- that is, by message number and UID. Therefore, the UID SEARCH, SEARCH, UID SORT, or SORT command used -- collectively known as the searching command -- will always have an order, the requested order, which will be the mailbox order for UID SEARCH and SEARCH commands.
All of the return specifiers have no interaction with either each other or any return specifiers defined in [ESEARCH] or Section 3.1; however, it is believed that implementations supporting CONTEXT will also support ESEARCH and ESORT.
The return option CONTEXT SHOULD be used by a client to indicate that subsequent use of the search criteria are likely. Servers MAY ignore this return option or use it as a hint to maintain a full result cache, or index.
Cridland & King Standards Track [Page 5]
RFC 5267 IMAP CONTEXT July 2008
A client might choose to obtain a count of matching messages prior to obtaining actual results. Here, the client signals its intention to fetch the results themselves:
The search return option UPDATE, if used by a client, causes the server to issue unsolicited notifications containing updates to the results that would be returned by an unmodified searching command. These update sets are carried in ADDTO and REMOVEFROM data items in ESEARCH responses.
These ESEARCH responses carry a search correlator of the searching command, hence clients MUST NOT reuse tags, as already specified in Section 2.2.1 of [IMAP]. An attempt to use UPDATE where a tag is already in use with a previous searching command that itself used UPDATE SHALL result in the server rejecting the searching command with a BAD response.
Both ADDTO and REMOVEFROM data items SHOULD be delivered to clients in a timely manner, as and when results change, whether by new messages arriving in the mailbox, metadata such as flags being changed, or messages being expunged.
Typically, this would occur at the same time as the FETCH, EXISTS, or EXPUNGE responses carrying the source of the change.
Updates will cease when the mailbox is no longer selected, or when the CANCELUPDATE command, defined in Section 4.3.5, is issued by the client, whichever is sooner.
Unlike [ACAP], there is no requirement that a context need be created with CONTEXT to use UPDATE, and in addition, the lack of UPDATE with a CONTEXT does not affect the results caused by later searching commands -- there is no snapshot facility.
There is no interaction between UPDATE and any other return options; therefore, use of RETURN (UPDATE MIN), for example, does not notify about the minimum UID or sequence number, but notifies instead about all changes to the set of matching messages. In particular, this means that a client using UPDATE and PARTIAL on the same search program could receive notifications about messages that do not currently interest it.
Cridland & King Standards Track [Page 6]
RFC 5267 IMAP CONTEXT July 2008
Finally, as specified in the errata to [IMAP], any message sequence numbers used in the search program are evaluated at the time the command is received; therefore, if the messages referred to by such message sequence numbers change, no notifications will be emitted.
This time, the client will require notifications of updates and chooses to obtain a count:
C: B01 UID SEARCH RETURN (UPDATE COUNT) DELETED KEYWORD $Junk S: * ESEARCH (TAG "B01") COUNT 74 S: B01 OK Search completed, will notify. // Note that the following is rejected, and has no effect: C: B01 SORT RETURN (UPDATE) FLAGGED S: B01 BAD Tag reuse
In some cases, the server MAY refuse to provide updates, such as if an internal limit on the number of update contexts is reached. In such a case, an untagged NO is generated during processing of the command with a response-code of NOUPDATE. The response-code contains, as argument, the tag of the search command for which the server is refusing to honour the UPDATE request.
Other return options specified SHALL still be honoured.
Servers MUST provide at least one updating context per client, and SHOULD provide more -- see Appendix B for strategies on reducing the impact of additional updating contexts. Since sorted contexts require a higher implementation cost than unsorted contexts, refusal to provide updates for a SORT command does not imply that SEARCH contexts will also be refused.
This time, the client will require notifications of updates, and chooses to obtain a count:
C: B02 UID SORT RETURN (UPDATE COUNT) UTF-8 KEYWORD $Junk S: * ESEARCH (TAG "B02") COUNT 74 S: * NO [NOUPDATE "B02"] Too many contexts S: B02 OK Search completed, will not notify.
Client handling might be to retry with a UID SEARCH command, or else cancel an existing context; see Section 4.3.5.
The result update set included in the return data item is specified as UIDs or message numbers, depending on how the UPDATE was specified. If the UPDATE was present in a SEARCH or SORT command, the results will be message numbers; in a UID SEARCH or UID SORT command, they will be UIDs.
The client MUST process ADDTO and REMOVEFROM return data items in the order they appear, including those within a single ESEARCH response. Correspondingly, servers MUST generate ADDTO and REMOVEFROM responses such that the results are maintained in the requested order.
As with any response aside from EXPUNGE, ESEARCH responses carrying ADDTO and/or REMOVEFROM return data items MAY be sent at any time. In particular, servers MAY send such responses when no command is in progress, during the processing of any command, or when the client is using the IDLE facility described in [IDLE]. Implementors are recommended to read [NOTIFY] as a mechanism for clients to signal servers that they are willing to process responses at any time, and are also recommended to pay close attention to Section 5.3 of [IMAP].
It is anticipated that typical server implementations will emit ADDTO when they normally emit the causal FETCH or EXISTS, and similarly emit REMOVEFROM when they normally emit the causal FETCH or EXPUNGE.
The ADDTO return data item contains, as payload, a list containing pairs of a context position and a set of result updates in the requested order to be inserted at the context position. Where the searching command is a SEARCH or UID SEARCH command, the context position MAY be zero. Each pair is processed in the order that it appears.
Note that an ADDTO containing message sequence numbers added as a result of those messages being delivered or appended MUST be sent after the EXISTS notification itself, in order that those sequence numbers are valid.
If the context position is non-zero, the result update is inserted at the given context position, meaning that the first result in the set will occupy the new context position after insertion, and any prior existing result at that context position will be shifted to a later context position.
Cridland & King Standards Track [Page 8]
RFC 5267 IMAP CONTEXT July 2008
Where the context position is zero, the client MAY insert the message numbers or UIDs in the result list such that the result list is maintained in mailbox order. In this case, servers are RECOMMENDED to order the result update into mailbox order to produce the shortest representation in set-syntax.
Note that this example assumes messages 23762 and 23763 with UIDs 32768 and 32769 (respectively) previously had neither \Deleted nor $Junk set. Also note that only the ADDTO is included, and not the (now changed) COUNT.
If the searching command "C01" initially generated a result list of 2734:2735, then the following three responses are equivalent, and yield a result list of 2731:2735:
The REMOVEFROM return data item contains, as payload, a list containing pairs of a context position and a set of result updates in the requested order to be removed starting from the context position. Where the searching command is a SEARCH or UID SEARCH command, the context position MAY be zero. Each pair is processed in the order that it appears.
If the context position is non-zero, the results are removed at the given context position, meaning that the first result in the set will occupy the given context position before removal, and any prior existing result at that context position will be shifted to an earlier context position.
Where the context position is zero, the client removes the message numbers or UIDs in the result list wherever they occur, and servers are RECOMMENDED to order the result list in mailbox order to obtain the best benefit from the set-syntax.
Cridland & King Standards Track [Page 9]
RFC 5267 IMAP CONTEXT July 2008
Note that a REMOVEFROM containing message sequence numbers removed as a result of those messages being expunged MUST be sent prior to the expunge notification itself, in order that those sequence numbers remain valid.
Here, a message in the result list is expunged. The REMOVEFROM is shown to happen without any command in progress; see Section 4.3.2. Note that EXPUNGE responses do not have this property.
When a client no longer wishes to receive updates, it may issue the CANCELUPDATE command, which will prevent all updates to the contexts named in the arguments from being transmitted by the server. The command takes, as arguments, one or more tags of the commands used to request updates.
The server MAY free any resource associated with a context so disabled -- however, the client is free to issue further searching commands with the same criteria and requested order, including PARTIAL requests.
C: B04 CANCELUPDATE "B01" S: B04 OK No further updates.
The PARTIAL search return option causes the server to provide in an ESEARCH response a subset of the results denoted by the sequence range given as the mandatory argument. The first result is 1; thus, the first 500 results would be obtained by a return option of "PARTIAL 1:500", and the second 500 by "PARTIAL 501:1000". This intentionally mirrors message sequence numbers.
A single command MUST NOT contain more than one PARTIAL or ALL search return option -- that is, either one PARTIAL, one ALL, or neither PARTIAL nor ALL is allowed.
For SEARCH results, the entire result list MUST be ordered in mailbox order, that is, in UID or message sequence number order.
Cridland & King Standards Track [Page 10]
RFC 5267 IMAP CONTEXT July 2008
Where a PARTIAL search return option references results that do not exist, by using a range which starts or ends higher than the current number of results, then the server returns the results that are in the set. This yields a PARTIAL return data item that has, as payload, the original range and a potentially missing set of results that may be shorter than the extent of the range.
Clients need not request PARTIAL results in any particular order. Because mailboxes may change, clients will often wish to use PARTIAL in combination with UPDATE, especially if the intent is to walk a large set of results; however, these return options do not interact -- the UPDATE will provide notifications for all matching results.
// Recall from A01 that there are 23764 results. C: A02 UID SEARCH RETURN (PARTIAL 23500:24000) UNDELETED UNKEYWORD $Junk C: A03 UID SEARCH RETURN (PARTIAL 1:500) UNDELETED UNKEYWORD $Junk C: A04 UID SEARCH RETURN (PARTIAL 24000:24500) UNDELETED UNKEYWORD $Junk S: * ESEARCH (TAG "A02") UID PARTIAL (23500:24000 ...) // 264 results in set syntax elided, // this spans the end of the results. S: A02 OK Completed. S: * ESEARCH (TAG "A03") UID PARTIAL (1:500 ...) // 500 results in set syntax elided. S: A03 OK Completed. S: * ESEARCH (TAG "A04") UID PARTIAL (24000:24500 NIL) // No results are present, this is beyond the end of the results. S: A04 OK Completed.
Server implementations MAY cache results from a SEARCH or SORT, whether or not hinted to by CONTEXT, in order to make subsequent searches more efficient, perhaps by recommencing a subsequent PARTIAL search where a previous search left off. However, servers MUST behave identically whether or not internal caching is taking place; therefore, any such cache is required to be updated as changes to the mailbox occur. An alternate strategy would be to discard results when any change occurs to the mailbox.
command-select =/ "CANCELUPDATE" 1*(SP quoted) ;; <command-select> from [IMAP]
context-position = number ;; Context position may be 0 for SEARCH result additions. ;; <number> from [IMAP]
modifier-context = "CONTEXT"
modifier-partial = "PARTIAL" SP partial-range
partial-range = nz-number ":" nz-number ;; A range 500:400 is the same as 400:500. ;; This is similar to <seq-range> from [IMAP], ;; but cannot contain "*".
modifier-update = "UPDATE"
search-return-opt =/ modifier-context / modifier-partial / modifier-update ;; All conform to <search-return-opt>, from [IMAP-ABNF]
resp-text-code =/ "NOUPDATE" SP quoted ;; <resp-text-code> from [IMAP]
This document defines additional IMAP4 capabilities. As such, it does not change the underlying security considerations of [IMAP]. The authors and reviewers believe that no new security issues are introduced with these additional IMAP4 capabilities.
Creation of a large number of contexts may provide an avenue for denial-of-service attacks by authorized users. Implementors may reduce this by limiting the number of contexts possible to create, via the protocol features described in Section 4.3.1; by reducing the impact of contexts by the implementation strategies described in Appendix B; and by logging context creation and usage so that administrative remedies may be applied.
Much of the design of this extension can be found in ACAP. Valuable comments, both in agreement and in dissent, were received from Alexey Melnikov, Arnt Gulbrandsen, Cyrus Daboo, Filip Navara, Mark Crispin, Peter Coates, Philip Van Hoof, Randall Gellens, Timo Sirainen, Zoltan
Cridland & King Standards Track [Page 13]
RFC 5267 IMAP CONTEXT July 2008
Ordogh, and others, and many of these comments have had significant influence on the design or the text. The authors are grateful to all those involved, including those not mentioned here.
[ABNF] Crocker, D. and P. Overell, "Augmented BNF for Syntax Specifications: ABNF", STD 68, RFC 5234, January 2008.
[ESEARCH] Melnikov, A. and D. Cridland, "IMAP4 Extension to SEARCH Command for Controlling What Kind of Information Is Returned", RFC 4731, November 2006.
[IMAP] Crispin, M., "INTERNET MESSAGE ACCESS PROTOCOL - VERSION 4rev1", RFC 3501, March 2003.
[IMAP-ABNF] Melnikov, A. and C. Daboo, "Collected Extensions to IMAP4 ABNF", RFC 4466, April 2006.
[KEYWORDS] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
[SORT] Crispin, M. and K. Murchison, "Internet Message Access Protocol - SORT and THREAD Extensions", RFC 5256, June 2008.
It is possible to use the facilities described within this memo to create a facility largely similar to a virtual mailbox, but handled on the client side.
Initially, the client SELECTs the real "backing" mailbox. Next, it can switch to a filtered view at any time by issuing a RETURN (COUNT UPDATE CONTEXT), and using RETURN (PARTIAL x:y) as the user scrolls, feeding the results into a FETCH as required to populate summary views.
A typically useful view is "UID SORT (DATE) RETURN (...) UTF-8 UNSEEN UNDELETED", which can be used to show the mailbox sorted into INTERNALDATE order, filtered to only show messages which are unread and not yet deleted.
Certain contexts are particularly useful for client developers wishing to present something similar to the common trash mailbox metaphor in limited bandwidth. The simple criteria of UNDELETED only matches undeleted messages, and the corresponding DELETED search key can be used to display a per-mailbox trash-like virtual mailbox.
The command "SEARCH RETURN (UPDATE) ALL" can be used to create a context that notifies immediately about expunged messages, yet will not affect message sequence numbers until the normal EXPUNGE message can be sent. This may be useful for clients wishing to show this behavior without losing the benefit of sequence numbering.
A client need not maintain any result cache at all, but instead it can maintain a simple count of messages matching the search criteria. Typically, this would use the SEARCH command, as opposed to UID SEARCH, due to its smaller representation. Such usage might prove useful in monitoring the number of flagged, but unanswered, messages, for example, with "SEARCH RETURN (UPDATE COUNT) FLAGGED UNANSWERED".
The creation of a context, and immediate access to it, can all be accomplished in a single round-trip. Therefore, whilst it is possible to elide resynchronization if no changes have occurred, it is simpler in most cases to resynchronize by simply recreating the context.
Although a server may cache the results, this is neither mandated nor required, especially when the client uses SEARCH or UID SEARCH commands. UPDATE processing, for example, can be achieved efficiently by comparison of the old flag state (if any) and the new, and PARTIAL can be achieved by re-running the search until the suitable window is required. This is a result of there being no snapshot facility.
For example, on a new message, the server can simply test for matches against all current UPDATE context search programs, and for any that match, send the ADDTO return data.
Similarly, for a flag change on an existing message, the server can check whether the message matched with its old flags, whether it matches with new flags, and provide ADDTO or REMOVEFROM return data accordingly if these results differ.
For PARTIAL requests, the server can perform a full search, discarding results until the lower bound is hit, and stopping the search when sufficient results have been obtained.
With some additional state, it is possible to restart PARTIAL searches, thus avoiding performing the initial discard phase.
For the best performance, however, caching the full search results is needed, which can allow for faster responses at the expense of memory. One reasonable strategy would be to balance this trade-off at run-time, discarding search results after a suitable timeout, and regenerating them as required.
This yields state requirements of storing the search program for any UPDATE contexts, and optionally storing both search program and (updated) results for further contexts as required.
Cridland & King Standards Track [Page 16]
RFC 5267 IMAP CONTEXT July 2008
Note that in the absence of a server-side results cache, it may be impossible to know if an expunged message previously matched unless the original message is still available. Therefore, some implementations may be forced into using a results cache in many circumstances.
UPDATE contexts created with SORT or UID SORT will almost certainly require some form of results caching, however.
Authors' Addresses
Dave Cridland Isode Limited 5 Castle Business Village 36, Station Road Hampton, Middlesex TW12 2BX GB
EMail: dave.cridland@isode.com
Curtis King Isode Limited 5 Castle Business Village 36, Station Road Hampton, Middlesex TW12 2BX GB
EMail: cking@mumbo.ca
Cridland & King Standards Track [Page 17]
RFC 5267 IMAP CONTEXT July 2008
Full Copyright Statement
Copyright (C) The IETF Trust (2008).
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.
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, THE IETF TRUST 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.
Intellectual Property
The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org.