RFC 5465

The IMAP NOTIFY Extension

๐Ÿ“… published February 2009, from ietf

RFC 5465: The IMAP NOTIFY Extension ๐ŸŽ€

okay bestie, this is the IMAP NOTIFY extension, and it's a whole vibe ๐Ÿฆ‹ Here are the title-block deets:

  • Number and title: RFC 5465, The IMAP NOTIFY Extension
  • Group: Network Working Group
  • Authors: A. Gulbrandsen (Oryx Mail Systems GmbH), C. King, and A. Melnikov (Isode Ltd.)
  • Category: Standards Track
  • Date: February 2009
  • Updates: 5267 (RFC 5267)

Status of This Memo โœจ

so this doc is an Internet standards track protocol for the Internet community ๐Ÿ’… It's also asking for discussion and suggestions for improvements. If you want the current standardization state and status of this protocol, check the current edition of the "Internet Official Protocol Standards" (STD 1). And the memo can be distributed freely, no limits ๐ŸŒธ

Copyright (c) 2009 IETF Trust and the persons identified as the document authors. All rights reserved.

This doc is under BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info), as they were on the publication date. Read them carefully, bestie, because they explain your rights and restrictions for this doc ๐Ÿซถ

Abstract ๐ŸŒˆ

this doc defines an IMAP extension that lets a client ask for specific kinds of unsolicited notifications about specific mailboxes, like when messages get added to or deleted from them ๐Ÿงš

Table of Contents ๐Ÿ“

1. Overview and Rationale ..........................................3
2. Conventions Used in This Document ...............................4
3. The NOTIFY Extension ............................................4
   3.1. The NOTIFY Command .........................................4
4. Interaction with the IDLE Command ...............................8
5. Event Types .....................................................8
   5.1. FlagChange and AnnotationChange ............................9
   5.2. MessageNew .................................................9
   5.3. MessageExpunge ............................................10
   5.4. MailboxName ...............................................11
   5.5. SubscriptionChange ........................................12
   5.6. MailboxMetadataChange .....................................12
   5.7. ServerMetadataChange ......................................13
   5.8. Notification Overflow .....................................13
   5.9. ACL (Access Control List) Changes .........................13
6. Mailbox Specification ..........................................14
   6.1. Mailbox Specifiers Affecting the Currently
        Selected Mailbox ..........................................14
   6.2. Personal ..................................................15
   6.3. Inboxes ...................................................15
   6.4. Subscribed ................................................15
   6.5. Subtree ...................................................15
   6.6. Mailboxes .................................................16
7. Extension to SEARCH and SORT Commands ..........................16
8. Formal Syntax ..................................................16
9. Security Considerations ........................................19
10. IANA Considerations ...........................................19
   10.1. Initial LIST-EXTENDED Extended Data Item Registrations ...19
11. Acknowledgements ..............................................20
12. Normative References ..........................................20
13. Informative References ........................................21

1. Overview and Rationale ๐Ÿฆ‹

okay so the IDLE command (defined in [RFC2177]) is how a client gets into a mode where the IMAP server pushes it notifications about IMAP mailstore events for the selected mailbox. Lowkey great, but IDLE doesn't restrict or control which server events can be sent, or what info the server sends back for each event. Also, IDLE only works on the selected mailbox, so you'd need an extra TCP connection for every mailbox ๐Ÿ™ˆ

This doc defines an IMAP extension that lets clients say what they're into when it comes to unsolicited events from the server. Clients only get the events they care about, and servers know they don't have to bother generating certain types of untagged responses. It's giving efficiency ๐Ÿ’…

Without the NOTIFY command from this doc, an IMAP server only tells the client about mailstore changes in these cases:

  • as the result of a client command (e.g., FETCH responses to a FETCH or STORE command),
  • as unsolicited responses sent just before the end of a command (e.g., EXISTS or EXPUNGE) as the result of changes in other sessions, and
  • during an IDLE command.

The NOTIFY command extends what info can come back in those last two cases. It also lets the server send info about updates between commands. Plus, it lets the client extend what gets sent unsolicited about the selected mailbox and ask for some update info about other mailboxes too โœจ

The way IDLE and NOTIFY work together is covered in Section 4.

For new messages that get delivered to or appended to the selected mailbox, NOTIFY can request that a set of attributes be sent to the client in an unsolicited FETCH response. This means a client can be a passive recipient of events and new mail, and stay fully synchronised without issuing any further commands except to change the state of the mailbox on the server. Main character energy ๐Ÿ‘‘

Some mobile clients, though, might only want mail "pushed" for mail that matches a SEARCH pattern. To cover that need, [RFC5267] is augmented by this doc, which extends the UPDATE return option so it can specify a list of fetch-atts to return when a new message is delivered or appended in another session ๐ŸŒธ

2. Conventions Used in This Document ๐ŸŽ€

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this doc are to be interpreted as described in [RFC2119]. Yes, those capitals are doing serious work, so don't skip them ๐Ÿ˜ญ

The acronym MSN stands for Message Sequence Numbers (see Section 2.3.1.2 of [RFC3501]).

Example lines prefaced by "C:" are sent by the client, and ones prefaced by "S:" are sent by the server. "[...]" means elision ๐Ÿงš

3. The NOTIFY Extension ๐Ÿ’–

IMAP servers that support this extension advertise the NOTIFY capability. This extension adds the NOTIFY command as defined in Section 5.1.

A server implementing this extension is not required to implement LIST-EXTENDED [RFC5258], even though a NOTIFY-compliant server must be able to return extended LIST responses, defined in [RFC5258]. So LIST-EXTENDED is optional for the server, but the extended LIST responses are a must for NOTIFY-compliant servers ๐Ÿซถ

3.1. The NOTIFY Command ๐Ÿ””

okay bestie, the NOTIFY command is literally the main event ๐Ÿ’… here's the shape of it:

   Arguments:  "SET"
               Optional STATUS indicator
               Mailboxes to be watched
               Events about which to notify the client

   Or
   Arguments:  "NONE"

   Responses:  Possibly untagged STATUS responses (for SET)

   Result:     OK - The server will notify the client as requested.
               NO -  Unsupported NOTIFY event, NOTIFY too complex or
                     expensive, etc.
               BAD - Command unknown, invalid, unsupported, or has
                     unknown arguments.

So the NOTIFY command tells the server that the client is always listening for event notifications, even when no command is in progress, and asks the server to notify it about a specific set of events ๐ŸŽ€ it comes in two forms. NOTIFY NONE means the client is not interested in any kind of event happening on the server. NOTIFY SET swaps out the current list of interesting events for a brand new one โœจ

Before NOTIFY gets used for the first time, the server only sends notifications while a command is being processed. Even then, it only tells the client about events on the selected mailbox (see Section 5 for definitions): MessageNew, MessageExpunge, or FlagChange. It does not notify the client about any events on other mailboxes, so it's giving very limited ๐Ÿ™ˆ

A successful NOTIFY command lasts until the next NOTIFY command or until the IMAP connection is closed.

A successful NOTIFY SET command MUST cause the server to immediately return any accumulated changes to the currently selected mailbox (if any), such as flag changes and new or expunged messages. So a successful NOTIFY SET implies an implicit NOOP command, which is honestly iconic ๐Ÿซถ

The NOTIFY SET command can request notifications of message-related changes to the selected mailbox, whatever that happens to be at the time the message notifications are generated. You do this by specifying either the SELECTED or the SELECTED-DELAYED mailbox selector (see Section 6.1) in the NOTIFY SET command. If the SELECTED/SELECTED-DELAYED mailbox selector is not specified in the NOTIFY SET command, that means the client doesn't want to receive any <message-event>s for the currently selected mailbox. This is the same as specifying SELECTED NONE ๐Ÿ™…โ€โ™€๏ธ

The client can also request notifications on other mailboxes, either by name or by a limited mailbox pattern match. Message-related notifications returned for the currently selected mailbox are the ones specified by the SELECTED/SELECTED-DELAYED mailbox specifier, even if the selected mailbox also shows up by name (or matches a pattern) in the command. Non-message-related notifications are controlled by mailbox specifiers other than SELECTED/SELECTED-DELAYED ๐ŸŒธ

If the NOTIFY command enables MessageNew, MessageExpunge, AnnotationChange, or FlagChange notifications for a mailbox other than the currently selected mailbox, and the client has specified the STATUS indicator parameter, then the server MUST send a STATUS response for that mailbox before NOTIFY's tagged OK. If MessageNew is enabled, the STATUS response MUST contain MESSAGES, UIDNEXT, and UIDVALIDITY. If MessageExpunge is enabled, the STATUS response MUST contain MESSAGES. If either AnnotationChange or FlagChange are included and the server also supports the CONDSTORE [RFC4551] and/or QRESYNC [RFC5162] extensions, the STATUS response MUST contain UIDVALIDITY and HIGHESTMODSEQ. Leaving out the STATUS indicator parameter lets the client skip those extra STATUS responses, which is useful if the client already grabbed this info before sending the NOTIFY command ๐Ÿงš

Clients are advised to limit the number of mailboxes used with NOTIFY. Especially if a client asks for events for all accessible mailboxes, the server may swamp the client with updates about shared mailboxes. That may reduce the client's battery life ๐Ÿ”‹ and it also wastes both server and network resources ๐Ÿ’€

For each mailbox specified, the server checks that the client has access using the following test:

  • If the name does not refer to an existing mailbox, the server MUST ignore it.
  • If the name refers to a mailbox that the client can't LIST, the server MUST ignore it. For a server that implements [RFC4314], this means that if the client doesn't have the 'l' (lookup) right for the name, then the server MUST ignore the mailbox. This behavior prevents disclosure of potentially confidential information to clients who don't have rights to know it ๐Ÿ”
  • If the name refers to a mailbox that the client can LIST (e.g., it has the 'l' right from [RFC4314]), but the client doesn't have another right required for processing of the specified event(s), then the server MUST respond with an untagged extended LIST response containing the \NoAccess name attribute.

The server SHOULD return the tagged OK response if the client has access to at least one of the mailboxes specified in the current list of interesting events. The server MAY return the tagged NO response if the client has no access to any of the specified mailboxes and no access can ever be granted in the future (e.g., the client specified an event for 'Subtree Bar/Foo', 'Bar/Foo' doesn't exist, and LIST returns \Noinferiors for the parent 'Bar') ๐Ÿฅฒ

If the notification would be prohibitively expensive for the server (e.g., "notify me of all flag changes in all mailboxes"), the server MAY refuse the command with a tagged NO [NOTIFICATIONOVERFLOW] response ๐Ÿšซ

If the client requests information for events of an unsupported type, the server MUST refuse the command with a tagged NO response (not a BAD). This response SHOULD contain the BADEVENT response code, which MUST list names of all events supported by the server.

Here's an example:

      S: * OK [CAPABILITY IMAP4REV1 NOTIFY]
      C: a login bob alice
      S: a OK Password matched
      C: b notify set status (selected MessageNew (uid
         body.peek[header.fields (from to subject)]) MessageExpunge)
         (subtree Lists MessageNew)
      S: * STATUS Lists/Lemonade (UIDVALIDITY 4 UIDNEXT 9999 MESSAGES
         500)
      S: [...]
      S: * STATUS Lists/Im2000 (UIDVALIDITY 901 UIDNEXT 1 MESSAGES 0)
      S: b OK done
      C: c select inbox
      S: [...] (the usual 7-8 responses to SELECT)
      S: c OK INBOX selected
            (Time passes.  A new message is delivered to mailbox
            Lists/Lemonade.)
      S: * STATUS Lists/Lemonade (UIDVALIDITY 4 UIDNEXT 10000
         MESSAGES 501)
            (Time passes.  A new message is delivered to inbox.)
      S: * 127 FETCH (UID 127001 BODY[HEADER.FIELDS (From To
         Subject)] {75}
      S: Subject: Re: good morning
      S: From: alice@example.org
      S: To: bob@example.org
      S:
      S: )
            (Time passes.  The client decides it wants to know about
            one more mailbox.  As the client already knows necessary
            STATUS information for all mailboxes below the Lists
            mailbox, and because "notify set status" would cause
            STATUS responses for *all* mailboxes specified in the
            NOTIFY command, including the ones for which the client
            already knows STATUS information, the client issues an
            explicit STATUS request for the mailbox to be added to
            the watch list, followed by the NOTIFY SET without the
            STATUS parameter.)
      C: d STATUS misc (UIDVALIDITY UIDNEXT MESSAGES)
      S: * STATUS misc (UIDVALIDITY 1 UIDNEXT 999)
      S: d STATUS completed
      C: e notify set (selected MessageNew (uid
         body.peek[header.fields (from to subject)]) MessageExpunge)
         (subtree Lists MessageNew) (mailboxes misc MessageNew)
      S: e OK done

4. Interaction with the IDLE Command โœจ

If IDLE [RFC2177] (as well as this extension) is supported, then while processing any IDLE command, the server MUST send exactly the same events as instructed by the client using the NOTIFY command.

NOTIFY makes IDLE unnecessary for some clients. If a client does not use MSNs and '*' in commands, it can request MessageExpunge and MessageNew for the selected mailbox by using the NOTIFY command instead of entering the IDLE mode ๐Ÿ™Œ

A client that uses MSNs and '*' in commands can still use the NOTIFY command if it specifies the SELECTED-DELAYED mailbox specifier in the NOTIFY command ๐Ÿ’–

5. Event Types ๐ŸŒˆ

Only some of the events in [RFC5423] can be expressed in IMAP, and for some of them there are several possible ways to express the event.

This section specifies the events of which an IMAP server can notify an IMAP client, and how.

The server SHOULD omit notifying the client if the event is caused by this client. For example, if the client issues CREATE and has requested a MailboxName event that would cover the newly created mailbox, the server SHOULD NOT notify the client of the MailboxName change. no need to hear about your own moves, bestie ๐Ÿ’…

All event types described in this document require the 'l' and 'r' rights (see [RFC4314]) on all observed mailboxes. Servers that don't implement [RFC4314] should map the above rights to their access-control model.

If the FlagChange and/or AnnotationChange events are specified, MessageNew and MessageExpunge MUST also be specified by the client. Otherwise, the server MUST respond with the tagged BAD response.

If one of MessageNew or MessageExpunge is specified, then both events MUST be specified. Otherwise, the server MUST respond with the tagged BAD response.

The client can tell the server not to send an event by leaving that event out of the list specified in NOTIFY SET, by using the NONE event specifier in the NOTIFY SET, or by using NOTIFY NONE. In particular, NOTIFY SET ... NONE can be used as a snapshot facility by clients ๐Ÿ“ธ

5.1. FlagChange and AnnotationChange ๐Ÿฆ‹

If the flag and/or message annotation change happens in the selected mailbox, the server MUST notify the client by sending an unsolicited FETCH response, which MUST include UID and FLAGS/ANNOTATION FETCH data items. It MAY also send new FLAGS and/or OK [PERMANENTFLAGS ...] responses.

If a search context is in effect as specified in [RFC5267], an ESEARCH ADDTO or ESEARCH REMOVEFROM will also be generated, if appropriate. In this case, the FETCH response MUST precede the ESEARCH response.

If the change happens in another mailbox, then the server responds with a STATUS response. The exact content of that STATUS response depends on various factors. If CONDSTORE [RFC4551] and/or QRESYNC [RFC5162] are enabled by the client, then the server sends a STATUS response that includes at least HIGHESTMODSEQ and UIDVALIDITY status data items. If the number of messages with the \Seen flag changes, the server MAY also include the UNSEEN data item in the STATUS response. If CONDSTORE/QRESYNC is not enabled by the client and the server chooses not to include the UNSEEN data item, the server does not notify the client. When this event is requested, the server MUST notify the client about mailbox UIDVALIDITY changes. This is done by sending a STATUS response that includes UIDVALIDITY.

FlagChange covers the MessageRead, MessageTrash, FlagsSet, and FlagsClear events in [RFC5423].

Example in the selected mailbox:

   S: * 99 FETCH (UID 9999 FLAGS ($Junk))

And in another mailbox, with CONDSTORE in use:

   S: * STATUS Lists/Lemonade (HIGHESTMODSEQ 65666665 UIDVALIDITY
   101)

5.2. MessageNew โœจ

Okay so this one covers both MessageNew and MessageAppend from [RFC5423] ๐Ÿ’Œ

If the new or appended message lands in the selected mailbox, the server tells the client with an unsolicited EXISTS response, then an unsolicited FETCH response with the info the client asked for ๐Ÿฆ‹. A FETCH response SHOULD NOT be generated for a new message the client made itself on this connection, like through an APPEND or COPY command it ran on the selected mailbox. The server MAY also send a RECENT response if it marks the message as \Recent ๐ŸŒธ

Note that one EXISTS response can cover several MessageAppend/MessageNew events, so don't panic if they're bundled ๐Ÿงš

If a search context is in effect as specified in [RFC5267], an ESEARCH ADDTO will also be generated when it's relevant ๐Ÿ” Here the EXISTS response MUST precede the ESEARCH response. Both the NOTIFY command and the SEARCH and SORT commands (see Section 7) can ask for attributes on new messages. Those attributes SHOULD be combined into a single FETCH response, and the server SHOULD avoid sending duplicate data ๐Ÿ’… The FETCH response(s) MUST follow any ESEARCH ADDTO responses.

If the new or appended message is in a different mailbox, the server sends an unsolicited STATUS (UIDNEXT MESSAGES) response for that mailbox ๐Ÿ“ฌ If the CONDSTORE extension [RFC4551] and/or the QRESYNC extension [RFC5162] is enabled, the HIGHESTMODSEQ status data item MUST be included in the STATUS response, no exceptions ๐Ÿ”ฅ

The client SHOULD NOT use FETCH attributes that implicitly set the \seen flag, or that assume a given bodypart exists ๐Ÿ™ˆ The most useful attributes are probably UID, MODSEQ, FLAGS, ENVELOPE, BODY.PEEK[HEADER.FIELDS... and BODY/BODYSTRUCTURE.

Here's the catch: if a client asks to be notified of MessageNew events with the SELECTED mailbox specifier, the number of messages can go up at any time. So the client can't refer to a specific message using the MSN/UID '*' ๐Ÿ˜ญ

Example in the selected mailbox:

S: * 444 EXISTS
S: * 444 FETCH (UID 9999)

And in another mailbox, without CONDSTORE enabled:

S: * STATUS Lists/Lemonade (UIDNEXT 10002 MESSAGES 503)

5.3. MessageExpunge ๐Ÿ—‘๏ธ

If the expunged message or messages are in the selected mailbox, the server tells the client using EXPUNGE (or VANISHED, if [RFC5162] is supported by the server and enabled by the client) ๐Ÿ’ซ

If a search context is in effect, as specified in [RFC5267], an ESEARCH REMOVEFROM will also be generated when it makes sense ๐Ÿ”

If the expunged message or messages are in another mailbox, the server sends an unsolicited STATUS (UIDNEXT MESSAGES) response for that mailbox. If the QRESYNC [RFC5162] extension is enabled, the HIGHESTMODSEQ data item MUST be included in the STATUS response as well ๐ŸŒˆ

Heads up: if a client requests MessageExpunge with the SELECTED mailbox specifier, the meaning of an MSN can change at any time, so the client can't use MSNs in commands anymore ๐Ÿ˜ต For example, that client can't use FETCH and has to use UID FETCH instead. The meaning of '*' can also shift when messages are added or expunged. A client that wants to keep using MSNs can either use the SELECTED-DELAYED mailbox specifier or skip the MessageExpunge event entirely ๐Ÿง 

The MessageExpunge notification covers both MessageExpunge and MessageExpire events from [RFC5423] ๐ŸŽ€

Example in the selected mailbox, without QRESYNC:

S: * 444 EXPUNGE

The same example in the selected mailbox, with QRESYNC:

S: * VANISHED 5444

And in another mailbox, when QRESYNC is not enabled:

S: * STATUS misc (UIDNEXT 999 MESSAGES 554)

5.4. MailboxName ๐Ÿ“

These notifications get sent when an affected mailbox name is created (with CREATE), deleted (with DELETE), or renamed (with RENAME) ๐Ÿ  For a server that implements [RFC4314], granting or revoking the 'l' right to the current user on the affected mailbox MUST be treated as mailbox creation or deletion, respectively. When a mailbox is created or deleted, the mailbox itself and its direct parent (whether or not that parent is an existing mailbox) count as affected ๐ŸŒธ

The server tells the client by sending an unsolicited LIST response for each affected mailbox name. If, after the event, the mailbox name doesn't refer to a mailbox the client can access, the \Nonexistent flag MUST be included ๐Ÿ‘€

For each LISTable mailbox that gets renamed, the server sends an extended LIST response [RFC5258] for the new name, containing the OLDNAME extended data item with the old mailbox name ๐Ÿฆ‹ When a mailbox is renamed, its children are renamed too, so no extra MailboxName events go out for them. When INBOX is renamed, a new INBOX is assumed to be created. And no MailboxName event is sent for INBOX in that case ๐Ÿ’…

If the server automatically subscribes a mailbox when it's created or renamed, then the unsolicited LIST response for each affected subscribed mailbox name MUST include the \Subscribed attribute (see [RFC5258]). The server SHOULD also include \HasChildren or \HasNoChildren attributes [RFC5258] when they apply โœจ

Example of a newly created mailbox (or granting of the 'l' right on the mailbox):

S: * LIST () "/" "NewMailbox"

And a deleted mailbox (or revocation of the 'l' right on the mailbox):

S: * LIST (\NonExistent) "." "INBOX.DeletedMailbox"

Example of a renamed mailbox:

S: * LIST () "/" "NewMailbox" ("OLDNAME" ("OldMailbox"))

5.5. SubscriptionChange ๐Ÿ“š

The server tells the client by sending an unsolicited LIST response for each affected mailbox name. If and only if the mailbox is subscribed after the event, the \Subscribed attribute (see [RFC5258]) is included ๐Ÿ’œ Note that in the LIST response, all mailbox attributes MUST be accurately computed. This is different from how the LSUB command behaves, so keep it in mind ๐Ÿ™ƒ

Example:

S: * LIST (\Subscribed) "/" "SubscribedMailbox"

5.6. MailboxMetadataChange ๐Ÿท๏ธ

Support for this event type is OPTIONAL, unless the METADATA extension [RFC5464] is also supported by the server. In that case, support for this event type is REQUIRED ๐Ÿ”’

A client that wants to receive unsolicited METADATA responses from the MailboxMetadataChange event in the NOTIFY command doesn't have to issue ENABLE METADATA ๐Ÿซถ

The server sends an unsolicited METADATA response (as per Section 4.4.2 of [RFC5464]). If possible, only the changed metadata SHOULD be included. But if the server can't detect a change to a single metadata item, it MAY include all metadata items set on the mailbox. If a metadata item is deleted (set to NIL), it MUST always be included in the METADATA response ๐Ÿ—‘๏ธ

Example:

S: * METADATA "INBOX" /shared/comment

5.7. ServerMetadataChange ๐Ÿ–ฅ๏ธ

Support for this event type is OPTIONAL, unless the METADATA or the METADATA-SERVER extension [RFC5464] is also supported by the server. In that case, support for this event type is REQUIRED ๐Ÿ’–

A client that wants unsolicited METADATA responses from the ServerMetadataChange event in the NOTIFY command doesn't have to issue ENABLE METADATA or ENABLE METADATA-SERVER ๐ŸŒท

The server sends an unsolicited METADATA response (as per Section 4.4.2 of [RFC5464]). Only the names of changed metadata entries SHOULD be returned in these METADATA responses. If a metadata item is deleted (set to NIL), it MUST always be included in the METADATA response ๐Ÿ’€

Example:

S: * METADATA "" /shared/comment

5.8. Notification Overflow ๐ŸŒŠ

If the server can't deliver as many notifications as it's being asked to, or isn't willing to, it may disable notifications for some or all clients. It MUST notify these clients by sending an untagged "OK [NOTIFICATIONOVERFLOW]" response, and then behave as if a NOTIFY NONE command had just been received ๐Ÿšจ

Example:

S: * OK [NOTIFICATIONOVERFLOW] ...A comment can go here...

5.9. ACL (Access Control List) Changes ๐Ÿ”

Even if NOTIFY succeeds, you can still lose access to the mailboxes being monitored later on ๐Ÿ˜ค If that happens, the server MUST stop monitoring those mailboxes. If access is granted again later, the server MUST restart event monitoring.

The server SHOULD return the LIST response with the \NoAccess name attribute when, and if, the mailbox loses the 'l' right. Similarly, the server SHOULD return the LIST response with no \NoAccess name attribute if the mailbox was previously reported as having \NoAccess and the 'l' right is later granted ๐Ÿฆ‹


6. Mailbox Specification ๐Ÿ“ฎ

Mailboxes to be monitored can be specified in a few different ways ๐Ÿ—‚๏ธ

Only 'SELECTED' and 'SELECTED-DELAYED' (Section 6.1) match the currently selected mailbox. Every other mailbox specification affects non-selected mailboxes ๐Ÿ‘‘

Multiple <event-group>s can apply to the same mailbox, and this example shows that ๐Ÿงš Here, MessageNew and MessageExpunge events get reported for INBOX because of the first <event-group>. A SubscriptionChange event also gets reported for INBOX, because of the second <event-group>.

C: a notify set (mailboxes INBOX (Messagenew messageExpunge))
   (personal (SubscriptionChange))

A typical client that supports the NOTIFY extension would ask for events on the selected mailbox and some named mailboxes ๐Ÿ’Œ

In the next example, the client asks for FlagChange events for all personal mailboxes except the currently selected mailbox. This is different from the previous example because SELECTED overrides all other message event definitions for the currently selected mailbox (see Section 3.1). Plot twist ๐Ÿ˜

C: a notify set (selected (Messagenew (uid flags) messageExpunge))
   (personal (MessageNew FlagChange MessageExpunge))

6.1. Mailbox Specifiers Affecting the Currently Selected Mailbox ๐ŸŽฏ

Only one of the mailbox specifiers affecting the currently selected mailbox can be specified in any NOTIFY command. The two such specifiers are SELECTED and SELECTED-DELAYED, and they're described below ๐Ÿ’…

Both refer to the mailbox that was selected using either SELECT or EXAMINE (see [RFC3501], Sections 6.3.1 and 6.3.2). When the IMAP connection isn't in the selected state, these mailbox specifiers don't refer to any mailbox ๐ŸŒ™

The mailbox specifiers only apply to <message-event>s. It's an error to specify other types of events with either the SELECTED or the SELECTED-DELAYED selector ๐Ÿšซ

6.1.1. Selected ๐Ÿ“

The SELECTED mailbox specifier requires the server to send immediate notifications for the currently selected mailbox about all specified <message-event>s โšก

6.1.2. Selected-Delayed โณ

The SELECTED-DELAYED mailbox specifier requires the server to delay a MessageExpunge event until the client issues a command that allows returning information about expunged messages (see Section 7.4.1 of [RFC3501] for more details). For example, that could be a NOOP or an IDLE command ๐ŸŒ™ When SELECTED-DELAYED is specified, the server MAY also delay returning other <message-event>s until the client issues one of the commands listed above, or it MAY return them immediately ๐Ÿคทโ€โ™€๏ธ

6.2. Personal ๐Ÿก

Personal refers to all selectable mailboxes in the user's personal namespace(s), as defined in [RFC2342] ๐ŸŽ€

6.3. Inboxes ๐Ÿ“ฌ

okay so inboxes means every selectable mailbox in the user's personal namespace(s) that a Message Delivery Agent (MDA) can deliver mail into (see [EMAIL-ARCH], particularly Section 4.3.3) ๐Ÿ’Œ

If the IMAP server can't easily work out that set, it MUST treat "inboxes" as equivalent to "personal". ๐Ÿซถ

6.4. Subscribed ๐Ÿ””

subscribed refers to every mailbox the user has subscribed to โœจ

If the subscription list changes, the server MUST reevaluate the list. ๐Ÿ”„

6.5. Subtree ๐ŸŒณ

subtree is followed by a mailbox name or a list of mailbox names ๐ŸŒธ A subtree covers all selectable mailboxes that sit underneath the specified mailbox, plus the specified mailbox itself. It's giving family tree energy ๐Ÿ‘‘

6.6. Mailboxes ๐Ÿ“

mailboxes is followed by a mailbox name or a list of mailbox names ๐Ÿ’œ The server MUST NOT do a wildcard expansion. This means there is no special treatment for the LIST wildcard characters ('*' and '%') if they show up in mailbox names. ngl that's a very clear boundary, and I respect it ๐Ÿ™ˆ


7. Extension to SEARCH and SORT Commands ๐Ÿ”

okay so if the server supports NOTIFY AND also supports CONTEXT=SEARCH and/or CONTEXT=SORT (both defined in [RFC5267]), the UPDATE return option gets extended. A client can ask for FETCH attributes to come back whenever a new message joins the context result set ๐Ÿฆ‹

Here's how it looks in practice:

      C: a00 SEARCH RETURN (COUNT UPDATE (UID BODY[HEADER.FIELDS (TO
         FROM SUBJECT)])) FROM "boss"
      S: * ESEARCH (TAG "a00") (COUNT 17)
      S: a00 OK
         [...a new message is delivered...]
      S: * EXISTS 93
      S: * 93 FETCH (UID 127001 BODY[HEADER.FIELDS (FROM TO SUBJECT)]
         {76}
      S: Subject: Re: good morning
      S: From: myboss@example.org
      S: To: bob@example.org
      S:
      S: )
      S: * ESEARCH (TAG "a00") ADDTO (0 93)

Note that the EXISTS response MUST precede any FETCH responses, and together they MUST precede the ESEARCH response. The order is strict, bestie, so don't mix it up ๐Ÿซฃ

No untagged FETCH response SHOULD be returned if a message becomes a member of UPDATE SEARCH because of flag or annotation changes. ๐Ÿšซ


8. Formal Syntax ๐Ÿ’…

The syntax below uses the Augmented Backus-Naur Form (ABNF) notation from [RFC5234]. [RFC3501] defines the non-terminals "capability", "command-auth", "mailbox", "mailbox-data", "resp-text-code", and "search-key". The "modifier-update" non-terminal is defined in [RFC5267]. "mbx-list-oflag" is defined in [RFC3501] and updated by [RFC5258] ๐Ÿงš

Except as noted otherwise, all alphabetic characters are case-insensitive. Using upper or lower case for token strings is only for editorial clarity, so it doesn't change anything ๐Ÿ’– Implementations MUST accept these strings in a case-insensitive fashion. For example, the <filter-mailboxes-selected> non-terminal value "SELECTED" must be treated in the same way as "Selected" or "selected". โœจ

   capability      =/ "NOTIFY"

   command-auth    =/ notify

   notify          = "NOTIFY" SP
                   (notify-set / notify-none)

   notify-set      = "SET" [status-indicator] SP event-groups
                   ; Replace registered notification events
                   ; with the specified list of events

   notify-none     = "NONE"
                   ; Cancel all registered notification
                   ; events.  The client is not interested
                   ; in receiving any events.

   status-indicator = SP "STATUS"

   one-or-more-mailbox = mailbox / many-mailboxes

   many-mailboxes  = "(" mailbox *(SP mailbox) ")"

   event-groups    = event-group *(SP event-group)

   event-group     = "(" filter-mailboxes SP events ")"
                   ;; Only <message-event>s are allowed in <events>
                   ;; when <filter-mailboxes-selected> is used.

   filter-mailboxes = filter-mailboxes-selected /
                   filter-mailboxes-other

   filter-mailboxes-other = "inboxes" / "personal" / "subscribed" /
                   ( "subtree" SP one-or-more-mailbox ) /
                   ( "mailboxes" SP one-or-more-mailbox )

   filter-mailboxes-selected = "selected" / "selected-delayed"
                   ;; Apply to the currently selected mailbox only.
                   ;; Only one of them can be specified in a NOTIFY
                   ;; command.

   events          = ( "(" event *(SP event) ")" ) / "NONE"
                   ;; As in [MSGEVENT].
                   ;; "NONE" means that the client does not wish
                   ;; to receive any events for the specified
                   ;; mailboxes.

   event           = message-event /
                   mailbox-event / user-event / event-ext

   message-event   = ( "MessageNew" [SP
                       "(" fetch-att *(SP fetch-att) ")" ] )
                   / "MessageExpunge"
                   / "FlagChange"
                   / "AnnotationChange"
                   ;; "MessageNew" includes "MessageAppend" from
                   ;; [MSGEVENT]. "FlagChange" is any of
                   ;; "MessageRead", "MessageTrash", "FlagsSet",
                   ;; "FlagsClear" [MSGEVENT]. "MessageExpunge"
                   ;; includes "MessageExpire" [MSGEVENT].
                   ;; MessageNew and MessageExpunge MUST always
                   ;; be specified together.  If FlagChange is
                   ;; specified, then MessageNew and MessageExpunge
                   ;; MUST be specified as well.
                   ;; The fett-att list may only be present for the
                   ;; SELECTED/SELECTED-DELAYED mailbox filter
                   ;; (<filter-mailboxes>).

   mailbox-event   = "MailboxName" /
                   "SubscriptionChange" / "MailboxMetadataChange"
                   ; "SubscriptionChange" includes
                   ; MailboxSubscribe and MailboxUnSubscribe.
                   ; "MailboxName" includes MailboxCreate,
                   ; "MailboxDelete" and "MailboxRename".

   user-event      = "ServerMetadataChange"

   event-ext       = atom
                   ;; For future extensions

   oldname-extended-item =  "OLDNAME" SP "(" mailbox ")"
                   ;; Extended data item (mbox-list-extended-item)
                   ;; returned in a LIST response when a mailbox is
                   ;; renamed.
                   ;; Note 1: the OLDNAME tag can be returned
                   ;; with or without surrounding quotes, as per
                   ;; mbox-list-extended-item-tag production.

   resp-text-code  =/ "NOTIFICATIONOVERFLOW" /
                   unsupported-events-code

   message-event-name   = "MessageNew" /
                   "MessageExpunge" / "FlagChange" /
                   "AnnotationChange"

   event-name = message-event-name / mailbox-event /
                   user-event

   unsupported-events-code = "BADEVENT"
                   SP "(" event-name *(SP event-name) ")"

   modifier-update = "UPDATE"
                   [ "(" fetch-att *(SP fetch-att) ")" ]

   mbx-list-oflag =/ "\NoAccess"

9. Security Considerations ๐Ÿšจ

It is very easy for a client to deny itself service using NOTIFY. Asking for all events on all mailboxes might work on a small server, but on a big server it can swamp the client's network connection or processing capability. In the worst case, the server's processing could also degrade the service it offers to other clients. ๐Ÿ˜ญ

Server authors, listen up: if a client issues requests and doesn't listen to the resulting responses, the TCP window can easily fill up, and a careless server might block. This problem also exists in plain IMAP, but this extension magnifies it ๐Ÿ’€

This extension makes it possible to retrieve messages immediately when they are added to the mailbox. That makes it wholly impractical to delete sensitive messages using programs like imapfilter. Using SIEVE [RFC5228] or similar is much better. ๐Ÿ”ฅ

10. IANA Considerations ๐Ÿงš

The IANA has added NOTIFY to the list of IMAP extensions ๐ŸŒˆ

10.1. Initial LIST-EXTENDED Extended Data Item Registrations ๐Ÿ“

The following entry has been added to the LIST-EXTENDED response registry [RFC5258]:

   To: iana@iana.org
   Subject: Registration of OLDNAME LIST-EXTENDED extended data item

   LIST-EXTENDED extended data item tag: OLDNAME

   LIST-EXTENDED extended data item description: The OLDNAME extended
      data item describes the old mailbox name for the mailbox
      identified by the LIST response.

   Which LIST-EXTENDED option(s) (and their types) causes this extended
      data item to be returned (if any): none

   Published specification : RFC 5465, Section 5.4.

   Security considerations: none

   Intended usage: COMMON

   Person and email address to contact for further information:  Alexey
      Melnikov <Alexey.Melnikov@isode.com>

   Owner/Change controller: iesg@ietf.org

11. Acknowledgments ๐Ÿซถ

The authors gratefully acknowledge the help of Peter Coates, Dave Cridland, Mark Crispin, Cyrus Daboo, Abhijit Menon-Sen, Timo Sirainen, and Eric Burger. In particular, Peter Coates contributed lots of text and useful suggestions to this document. The whole group chat is thanking Peter rn ๐Ÿ’œ

Various examples are copied from other RFCs. ๐Ÿ“š

This document builds on one published and two unpublished drafts by the same authors. โœจ


12. Normative References ๐Ÿ“–

   [RFC2119]    Bradner, S., "Key words for use in RFCs to Indicate
                Requirement Levels", BCP 14, RFC 2119, March 1997.

   [RFC2177]    Leiba, B., "IMAP4 IDLE command", RFC 2177, June 1997.

   [RFC2342]    Gahrns, M. and C. Newman, "IMAP4 Namespace", RFC 2342,
                May 1998.

   [RFC3501]    Crispin, M., "INTERNET MESSAGE ACCESS PROTOCOL - VERSION
                4rev1", RFC 3501, March 2003.

   [RFC4314]    Melnikov, A., "IMAP4 Access Control List (ACL)
                Extension", RFC 4314, December 2005.

   [RFC4466]    Melnikov, A. and C. Daboo, "Collected Extensions to
                IMAP4 ABNF", RFC 4466, April 2006.

   [RFC4551]    Melnikov, A. and S. Hole, "IMAP Extension for
                Conditional STORE Operation or Quick Flag Changes
                Resynchronization", RFC 4551, June 2006.

   [RFC5162]    Melnikov, A., Cridland, D., and C. Wilson, "IMAP4
                Extensions for Quick Mailbox Resynchronization", RFC
                5162, March 2008.

   [RFC5234]    Crocker, D., Ed., and P. Overell, "Augmented BNF for
                Syntax Specifications: ABNF", STD 68, RFC 5234, January
                2008.

   [RFC5258]    Leiba, B. and A. Melnikov, "Internet Message Access
                Protocol version 4 - LIST Command Extensions", RFC 5258,
                June 2008.

   [RFC5267]    Cridland, D. and C. King, "Contexts for IMAP4", RFC
                5267, July 2008.

   [RFC5423]    Newman, C. and R. Gellens, "Internet Message Store
                Events", RFC 5423, Month 2009.

   [RFC5464]    Daboo, C., "The IMAP METADATA Extension", RFC 5464,
                February 2009.

13. Informative References ๐ŸŽ€

   [RFC5228]    Guenther, P., Ed., and T. Showalter, Ed., "Sieve: An
                Email Filtering Language", RFC 5228, January 2008.

   [EMAIL-ARCH] Crocker, D., "Internet Mail Architecture", Work in
                Progress, October 2008.

Authors' Addresses ๐Ÿ’Œ

   Arnt Gulbrandsen
   Oryx Mail Systems GmbH
   Schweppermannstr. 8
   D-81671 Muenchen
   Germany

   EMail: arnt@oryx.com


   Curtis King
   Isode Ltd
   5 Castle Business Village
   36 Station Road
   Hampton, Middlesex  TW12 2BX
   UK

   EMail: Curtis.King@isode.com


   Alexey Melnikov
   Isode Ltd
   5 Castle Business Village
   36 Station Road
   Hampton, Middlesex  TW12 2BX
   UK

   EMail: Alexey.Melnikov@isode.com

That's the whole RFC, bestie! ๐ŸŒˆ๐Ÿ’–๐Ÿฆ‹โœจ

๐Ÿ“ฆ where this page comes from